4 team plans · 9 tasks · 40 models · Plan prices read on November 16, 2026 (Microsoft: October 25, 2026) · Scores from November 16, 2026
What the AI apps cost a team, per person per month
Plan
Price per person
ChatGPT Business (standard seat)2 to 200 people. A premium seat, with 5 times the use, costs $125 a month, or $100 a month paid for a year.
$25 a monthOr $20 a month, paid for a year. SourceFrom Hungary: HUF 7,950 a month. Or HUF 6,350 a month, paid for a year.
Claude Team (standard seat)2 to 150 people. A premium seat, with 5 times the use, costs $125 a month, or $100 a month paid for a year.
$25 a monthOr $20 a month, paid for a year. SourceFrom Hungary: €21.04 a month. Or €18 a month, paid for a year.
Google Workspace Business StandardUp to 300 people. The Gemini assistant in Gmail, Docs, Meet and more. Business Starter, at $8.40 a month ($7 for a year), has it in Gmail, not in Docs or Meet.
$16.80 a monthOr $14 a month, paid for a year. SourceFrom Hungary: €13.60 a month. Monthly; 16% off for a year, no price shown. New customers: 20% off the first 3 months, for a limited time.
Microsoft 365 Copilot Business (an add-on)Up to 300 people. Only as an add-on to a Microsoft 365 Business plan that you already pay for. Some advanced AI features cost extra, by use.
$18.00 a monthPaid for a year: an offer until December 31, 2026 (normally $21.00). Or $25.20 a month, paid monthly. Source
What your team pays, in US dollars
Plan
Paid monthly
Paid for a year
ChatGPT Business
$250 a month
$2,400 a year
Claude Team
$250 a month
$2,400 a year
Google Workspace Business Standard
$168 a month
$1,680 a year
Microsoft 365 Copilot Business
$252 a month
$2,160 a year
Worked out on this page: we don’t send or save your number. Before tax. Microsoft’s yearly price is an offer until December 31, 2026, and you also pay for the Microsoft 365 plan it adds to.
From each maker’s own price page or help center, read on November 16, 2026 (Microsoft: October 25, 2026). Check the page before you pay. This is a list of prices, not advice on what to buy.
Free plans
ChatGPT FreeUnlimited everyday text chats, web search, file uploads and making images, with limits on the tools. Ads may show in some countries.
Claude FreeChat on the web, computer and phone, web search, files and code, memory, and your connected apps, with less use than Pro.
Gemini (free)A lighter model (Flash-Lite), making and editing images, Deep Research, voice chats (Gemini Live), and 15 GB of storage, with a Google account.
ChatGPT Plus$20 a monthMonthly only. SourceFrom Hungary: HUF 8,990 a month. Monthly.ChatGPT for one person.
Claude Pro$20 a month, or $17 a month for a yearThe year is $200, paid up front. SourceFrom Hungary: €18 a month, or €15 a month for a year. Monthly, or yearly.Claude for one person.
Google AI Pro$19.99 a monthMonthly. SourceFrom Hungary: HUF 8,790 a month. Monthly.Gemini for one person, shared with up to 5 family members if you like. Businesses use Workspace instead.
Big companies get prices on request: ChatGPT Enterprise and Google Workspace Enterprise show none. Claude Enterprise costs $20 a seat a month, paid for a year, for 20 seats or more, plus what the team uses at API prices.
Prices are without tax. The first price is the maker’s US-dollar list price; the second is what its page showed from Hungary. Your country may see another currency or price: the plan’s name links to its page.
Our pick for each task
9 tasks
Pick a task for our pick, the other choices and the top ten. Rankings come from public tests whose data may be reused.
Which model is best for everyday questions and writing?
Used for: emails, texts, reports, captions.
Our pickGemini 4 CarbonGoogle · for developers: $2 in · $10 outUntil November 30, then $4 in · $20 outThe best model anyone can use, with enough evidence.
Best you can downloadKimi K3Moonshot AI (Kimi) · No public API price · downloadOpen weights: you can run it on your own computers.
Cheaper choiceGemini 3.8 FlashGoogle · for developers: $0.75 in · $3.75 outUntil December 31, then $1.50 in · $7.50 outIn the top five, about 63% cheaper than our pick, and 7.8 points behind it.
Place
Model
Score
Price for developers
1
Gemini 4 CarbonGoogle
100.0
$2 in · $10 outUntil November 30, then $4 in · $20 out
2
Gemini 4 ArgonGoogle
98.5
$2 in · $10 outUntil November 30, then $4 in · $20 out
3
Claude Fable 5Anthropic
93.3
$10 in · $50 out
4
Gemini 3.8 FlashGoogle
92.2
$0.75 in · $3.75 outUntil December 31, then $1.50 in · $7.50 out
5
Claude Fable 5.1Anthropic
92.0
$10 in · $50 out
6
GPT-6.1 SolOpenAI
91.0
$2 in · $10 out
7
Gemini 3.7 FlashGoogle
90.6
No public API price
8
GPT-6 AstraOpenAI
90.6
$10 in · $50 out
9
Gemini 3.1 ProGoogle
90.0
$2 in · $12 out
10
Muse Spark 1.3Meta
89.5
No public API price
Each model has its own place, even when scores are close. Scores a few points apart are about as good, because tests often disagree. Prices are what developers pay to use a model through its API, in US dollars per million tokens (roughly 750,000 English words). In the apps, most people use a free or a monthly plan instead: see Plans and prices. From each company’s own price page.
Ranked by our Silicon score, from 6 tests by 4 publishers. Scores from November 16, 2026.
Answering questions, writing and everyday help, as in ChatGPT, Claude or Gemini. Tests: Arena Text, LiveBench Language, LiveBench Instruction Following, SimpleQA Verified, FACTS Grounding, FACTS Parametric. How the Silicon score works.
Coding
Which model is best at coding?
Used for: websites, apps and small programs.
Our pickClaude Opus 5.5Anthropic · for developers: $4 in · $20 outThe best model anyone can use, with enough evidence.
Best you can downloadDeepSeek V4.1 FlashDeepSeek · for developers: $0.30 in · $1.20 out · downloadOpen weights: you can run it on your own computers.
Cheaper choiceGPT-6.1 TerraOpenAI · for developers: $1 in · $5 outIn the top five, about 75% cheaper than our pick, and 11.7 points behind it.
Place
Model
Score
Price for developers
1
Claude Opus 5.5Anthropic
100.0
$4 in · $20 out
2
Claude Sonnet 5.5Anthropic
95.8
$2 in · $10 out
3
Claude Fable 5.1Anthropic
94.6
$10 in · $50 out
4
Claude Opus 5Anthropic
89.8
$5 in · $25 out
5
GPT-6.1 TerraOpenAI
88.3
$1 in · $5 out
6
DeepSeek V4.1 FlashDeepSeek · download
87.7
$0.30 in · $1.20 out
7
GPT-6 AstraOpenAI
86.0
$10 in · $50 out
8
Claude Fable 5Anthropic
85.8
$10 in · $50 out
9
GPT-6.1 SolOpenAI
85.5
$2 in · $10 out
10
Gemini 4 ArgonGoogle
84.4
$2 in · $10 outUntil November 30, then $4 in · $20 out
Each model has its own place, even when scores are close. Scores a few points apart are about as good, because tests often disagree. Prices are what developers pay to use a model through its API, in US dollars per million tokens (roughly 750,000 English words). In the apps, most people use a free or a monthly plan instead: see Plans and prices. From each company’s own price page.
Ranked by our Silicon score, from 4 tests by 3 publishers. Scores from November 16, 2026.
Writing code, and building websites and apps from a short description. Tests: Arena WebDev, MirrorCode, LiveBench Coding, LiveBench Agentic Coding. How the Silicon score works.
Math & reasoning
Which model is best at math and hard problems?
Used for: sums, budgets and step-by-step problems.
Our pickGemini 4 CarbonGoogle · for developers: $2 in · $10 outUntil November 30, then $4 in · $20 outThe best model anyone can use, with enough evidence.
Best you can downloadDeepSeek V4.1 FlashDeepSeek · for developers: $0.30 in · $1.20 out · downloadOpen weights: you can run it on your own computers.
Place
Model
Score
Price for developers
1
Gemini 4 CarbonGoogle
100.0
$2 in · $10 outUntil November 30, then $4 in · $20 out
2
GPT-6.1 SolOpenAI
97.9
$2 in · $10 out
3
GPT-6 AstraOpenAI
97.2
$10 in · $50 out
4
Claude Opus 5.5Anthropic
94.4
$4 in · $20 out
5
Claude Sonnet 5.5Anthropic
93.3
$2 in · $10 out
6
Claude Fable 5.1Anthropic
92.1
$10 in · $50 out
7
GPT-5.6 SolOpenAI
91.9
$4 in · $20 out
8
Claude Opus 5Anthropic
90.2
$5 in · $25 out
9
Gemini 4 ArgonGoogle
90.1
$2 in · $10 outUntil November 30, then $4 in · $20 out
10
GPT-6 SolOpenAI
89.4
$2 in · $10 out
Each model has its own place, even when scores are close. Scores a few points apart are about as good, because tests often disagree. Prices are what developers pay to use a model through its API, in US dollars per million tokens (roughly 750,000 English words). In the apps, most people use a free or a monthly plan instead: see Plans and prices. From each company’s own price page.
Ranked by our Silicon score, from 8 tests by 3 publishers. Scores from November 16, 2026.
Solving hard math problems that need many careful steps. Tests: FrontierMath Tiers 1–3, Mystery Game Puzzles, GPQA Diamond, Chess Puzzles, ArXivMath, BrokenArXiv, LiveBench Mathematics, LiveBench Reasoning. How the Silicon score works.
AI agents
Which model works best as an AI agent?
Used for: longer jobs done for you, step by step.
Our pickClaude Fable 5.2Anthropic · for developers: $8 in · $40 outThe best model anyone can use, with enough evidence.
Best you can downloadKimi K3Moonshot AI (Kimi) · No public API price · downloadOpen weights: you can run it on your own computers.
Cheaper choiceGPT-6.1 TerraOpenAI · for developers: $1 in · $5 outIn the top five, about 88% cheaper than our pick, and 17.3 points behind it.
Place
Model
Score
Price for developers
1
Claude Fable 5.2Anthropic
100.0
$8 in · $40 out
2
Gemini 4 CarbonGoogle
88.6
$2 in · $10 outUntil November 30, then $4 in · $20 out
3
Claude Opus 5.5Anthropic
88.0
$4 in · $20 out
4
GPT-6 AstraOpenAI
84.8
$10 in · $50 out
5
GPT-6.1 TerraOpenAI
82.7
$1 in · $5 out
6
Claude Fable 5.1Anthropic
82.4
$10 in · $50 out
7
GPT-6.1 SolOpenAI
80.0
$2 in · $10 out
8
Gemini 4 ArgonGoogle
79.7
$2 in · $10 outUntil November 30, then $4 in · $20 out
9
Claude Opus 5Anthropic
74.9
$5 in · $25 out
10
GPT-6 SolOpenAI
74.8
$2 in · $10 out
Each model has its own place, even when scores are close. Scores a few points apart are about as good, because tests often disagree. Prices are what developers pay to use a model through its API, in US dollars per million tokens (roughly 750,000 English words). In the apps, most people use a free or a monthly plan instead: see Plans and prices. From each company’s own price page.
Ranked by our Silicon score, from 3 tests by 3 publishers. Scores from November 16, 2026.
Doing longer jobs on its own, such as research or fixing code, using tools along the way. Tests: Arena Agent, EBR-bench, τ³-bench Banking. How the Silicon score works.
Web research
Which model finds the best answers on the web?
Used for: finding facts and sources online.
Our pickGPT-5.6 SolOpenAI · for developers: $4 in · $20 outThe best model anyone can use, with enough evidence.
Place
Model
Score
Price for developers
1
GPT-5.6 SolOpenAI
100.0
$4 in · $20 out
2
GPT-6 AstraOpenAI
96.9
$10 in · $50 out
3
GPT-5.5OpenAI
96.0
$5 in · $30 out
4
Qwen3.8 MaxAlibaba (Qwen)
96.0
No public API price
5
Claude Fable 5.1Anthropic
95.6
$10 in · $50 out
6
GPT-6.1 SolOpenAI
95.5
$2 in · $10 out
7
Claude Opus 4.6Anthropic
94.0
$5 in · $25 out
8
Claude Opus 5Anthropic
93.7
$5 in · $25 out
9
Claude Fable 5Anthropic
93.3
$10 in · $50 out
10
Claude Opus 5.5Anthropic
91.1
$4 in · $20 out
Each model has its own place, even when scores are close. Scores a few points apart are about as good, because tests often disagree. Prices are what developers pay to use a model through its API, in US dollars per million tokens (roughly 750,000 English words). In the apps, most people use a free or a monthly plan instead: see Plans and prices. From each company’s own price page.
Ranked by our Silicon score, from 3 tests by 3 publishers. Scores from November 16, 2026.
Searching the web, reading what it finds and answering with sources. Tests: SimpleQA Verified, Arena Search, FACTS Search. How the Silicon score works.
Understanding images
Which model understands images and documents best?
Used for: reading photos, receipts and screenshots.
Our pickClaude Opus 5.5Anthropic · for developers: $4 in · $20 outThe best model anyone can use, with enough evidence.
Best you can downloadDeepSeek V4.1 FlashDeepSeek · for developers: $0.30 in · $1.20 out · downloadOpen weights: you can run it on your own computers.
Cheaper choiceGPT-6.1 SolOpenAI · for developers: $2 in · $10 outIn the top five, about 50% cheaper than our pick, and 1.1 points behind it.
Place
Model
Score
Price for developers
1
Claude Opus 5.5Anthropic
100.0
$4 in · $20 out
2
Claude Fable 5.1Anthropic
99.6
$10 in · $50 out
3
GPT-6.1 SolOpenAI
98.9
$2 in · $10 out
4
Claude Opus 5Anthropic
96.7
$5 in · $25 out
5
Claude Fable 5Anthropic
95.5
$10 in · $50 out
6
GPT-6 AstraOpenAI
94.6
$10 in · $50 out
7
Gemini 3.8 FlashGoogle
93.1
$0.75 in · $3.75 outUntil December 31, then $1.50 in · $7.50 out
8
GPT-5.6 SolOpenAI
92.6
$4 in · $20 out
9
Qwen3.8 MaxAlibaba (Qwen)
92.0
No public API price
10
Gemini 3.7 FlashGoogle
91.7
No public API price
Each model has its own place, even when scores are close. Scores a few points apart are about as good, because tests often disagree. Prices are what developers pay to use a model through its API, in US dollars per million tokens (roughly 750,000 English words). In the apps, most people use a free or a monthly plan instead: see Plans and prices. From each company’s own price page.
Ranked by our Silicon score, from 4 tests by 3 publishers. Scores from November 16, 2026.
Reading photos, screenshots, charts and documents, and answering questions about them. Tests: Arena Vision, Arena Document, Furniture Assembly, FACTS Multimodal. How the Silicon score works.
Making images
Which model makes the images people prefer?
Used for: pictures for posts and ads.
Our pickGPT Image 2.5 SunburstOpenAI · Priced per imageThe best model anyone can use, with enough evidence.
Best you can downloadIdeogram 4.0 (quality)Ideogram · Priced per image · downloadOpen weights: you can run it on your own computers.
Place
Model
Score
Price for developers
1
GPT Image 2.5 SunburstOpenAI
1,429
Priced per image
=2
GPT Image 2.6 Sunburst (medium)OpenAI
1,406
Priced per image
=2
GPT Image 2.5 FlareOpenAI
1,398
Priced per image
=2
GPT Image 2.6 Flare (medium)OpenAI
1,384
Priced per image
=5
MAI Image 2.6Microsoft
1,330
Priced per image
=5
Nano Banana 2.1Google
1,327
Priced per image
7
Reve 2.1Reve
1,301
Priced per image
8
Grok Imagine Image 2.0xAI
1,287
Priced per image
9
Muse ImageMeta
1,266
Priced per image
=10
Nano Banana 2Google
1,256
Priced per image
Places count only the models we follow; the full Arena Text-to-Image list has more. Models with overlapping uncertainty share a place (shown as “=”). Image models charge per image, and each company counts it in its own way, so we show no price here. Most people make pictures in an app, with a free or a monthly plan: see Plans and prices.
Ranked by Arena Text-to-Image (Arena (LMArena)). Scores from November 12, 2026.
Making pictures from a written description. Source: Arena Text-to-Image by Arena (LMArena), November 12, 2026, licence CC BY 4.0.
Making videos
Which model makes the videos people prefer?
Used for: short clips for social media and ads.
Our pickGemini Omni 1.2 FlashGoogle · Priced per secondThe best model anyone can use, with enough evidence.
Just as goodGemini Omni 1.1 FlashGoogle · Priced per secondIts score is within the uncertainty of our pick.
Best you can downloadMiniMax H3MiniMax · Priced per second · downloadOpen weights: you can run it on your own computers.
Place
Model
Score
Price for developers
=1
Gemini Omni 1.2 FlashGoogle
1,540
Priced per second
=1
Gemini Omni 1.1 FlashGoogle
1,513
Priced per second
=3
FLUX 3 VideoBlack Forest Labs
1,491
Priced per second
=3
Grok Imagine Video 1.5xAI
1,488
Priced per second
=3
Seedance 2.0 (720p)ByteDance
1,483
Priced per second
=3
Wan 3.0Alibaba (Qwen)
1,476
Priced per second
=7
Muse VideoMeta
1,457
Priced per second
=7
MiniMax H3MiniMax · download
1,456
Priced per second
=9
Runway Gen-4.5Runway
1,223
Priced per second
–No place yet
Sora 2 ProOpenAI · API shut down
1,367
No API since September 24
Places count only the models we follow; the full Arena Text-to-Video list has more. Models with overlapping uncertainty share a place (shown as “=”). Models with few tests yet, for testers only, or without an API have no place (“–”) and come after the others. Video models charge per second of video, not per token, so we show no price here. Most people make videos in an app, with a free or a monthly plan: see Plans and prices.
Ranked by Arena Text-to-Video (Arena (LMArena)). Scores from November 9, 2026.
Making short video clips from a written description. Source: Arena Text-to-Video by Arena (LMArena), November 9, 2026, licence CC BY 4.0.
Search & retrieval
Which model is best at turning text into search-ready numbers?
Used for: search in your own files (for developers).
Our pickHarrier OSS v1 27BMicrosoft · Price not listed · downloadThe best model anyone can use, with enough evidence.
Best you can downloadHarrier OSS v1 27BMicrosoft · Price not listed · downloadOpen weights: you can run it on your own computers.
Place
Model
Score
Price for developers
1
Harrier OSS v1 27BMicrosoft · download
74.3
Price not listed
2
KaLM Embedding Gemma3 12BTencent · download
72.3
Price not listed
3
Qwen3 Embedding 8BAlibaba (Qwen) · download
70.6
Price not listed
4
Seed1.6 EmbeddingByteDance
70.3
Price not listed
5
Llama Embed Nemotron 8BNVIDIA · download
69.5
Price not listed
6
F2LLM v2 14BAnt Group (CodeFuse) · download
68.7
Price not listed
7
Gemini Embedding 001Google
68.4
Price not listed
8
Octen Embedding 8BOcten · download
67.8
Price not listed
9
Jina Embeddings v5 Omni SmallJina AI · download
67.0
Price not listed
Places count only the models we follow; the full MTEB Multilingual v2 list has more. We don’t list prices for these models yet: see each maker’s own page. You can run the ones marked “download” on your own computers.
Ranked by MTEB Multilingual v2 (MTEB). Scores from November 9, 2026.
Helping search tools find the right documents. This one is mostly for developers. Source: MTEB Multilingual v2 by MTEB, November 9, 2026, licence CC0.
Every model, one score
40 models
One score for 6 jobs, each counted the same, and each company’s official price for developers. Pick a model to see how its score is made.
Scores from November 16, 2026. Developer prices read on November 16, 2026.
The top score is 100. Scores a few points apart are close, because tests often disagree. A model with few tests yet (scored in fewer than 3 of the 6 jobs, or on too few tests) is listed without a place: its score may still move. Prices are what developers pay: official list prices in US dollars per million tokens, input and output. A team that uses the apps pays per person instead: see Plans and prices.
Price for developers$8 / $40in / out, per million tokens
Each job on its own · scored in 1 of 6
AI agents100.0
Its tests
Claude Fable 5.2 took 9 of our tests, from 2 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
94%
1 of 35
26%
Arena AgentArena (LMArena)
+28.61 points
1 of 39
19%
Arena WebDevArena (LMArena)
1826 rating
1 of 48
16%
Furniture AssemblyEpoch AI
90%
1 of 31
12%
EBR-benchEpoch AI
90.48%
1 of 21
10%
FrontierMath Tiers 1–3Epoch AI
95.09%
1 of 38
5%
Mystery Game PuzzlesEpoch AI
90%
1 of 36
5%
GPQA DiamondEpoch AI
93.88%
12 of 35
3%
Chess PuzzlesEpoch AI
79%
1 of 37
3%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it. Few tests yet: it has a job score of its own in only 1 of the 6 jobs, so the score may still move. It is listed, but gets no place.
GoogleClosed model
Gemini 4 Carbon
Place#1of 35 models with enough tests
Silicon score92.589.9 to 95.1 with other tests
Price for developers$2 / $10in / out, per million tokens, until November 30. Then $4 / $20.
Each job on its own · scored in 3 of 6
Chat & writing100.0
Math & reasoning100.0
AI agents88.6
Its tests
Gemini 4 Carbon took 12 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
77.7%
2 of 35
20%
Arena AgentArena (LMArena)
+16.45 points
3 of 39
15%
Arena WebDevArena (LMArena)
1768 rating
4 of 48
12%
Arena TextArena (LMArena)
1526 rating
1 of 41
11%
Furniture AssemblyEpoch AI
77.5%
6 of 31
9%
EBR-benchEpoch AI
87.62%
2 of 21
7%
ArXivMathMathArena (ETH Zürich)
94.28%
1 of 7
6%
BrokenArXivMathArena (ETH Zürich)
84.4%
1 of 7
6%
FrontierMath Tiers 1–3Epoch AI
90.53%
6 of 38
4%
Mystery Game PuzzlesEpoch AI
73%
5 of 36
4%
GPQA DiamondEpoch AI
96.53%
1 of 35
2%
Chess PuzzlesEpoch AI
78%
2 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Opus 5.5
Place#2of 35 models with enough tests
Silicon score87.385.5 to 89.1 with other tests
Price for developers$4 / $20in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing89.2
Coding100.0
Math & reasoning94.4
AI agents88.0
Web research91.1
Understanding images100.0
Its tests
Claude Opus 5.5 took 25 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
72.2%
7 of 35
10%
Arena SearchArena (LMArena)
1213 rating
12 of 17
9%
Arena VisionArena (LMArena)
1295 rating
5 of 30
9%
Arena AgentArena (LMArena)
+18.19 points
2 of 39
7%
Arena WebDevArena (LMArena)
1816 rating
2 of 48
6%
MirrorCodeEpoch AI
95.56%
2 of 32
6%
Arena TextArena (LMArena)
1507 rating
4 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
86.52%
4 of 13
4%
LiveBench ReasoningLiveBench
92.15%
3 of 34
4%
EBR-benchEpoch AI
71.43%
5 of 21
4%
Furniture AssemblyEpoch AI
83.33%
2 of 31
4%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
48.37%
4 of 13
4%
LiveBench LanguageLiveBench
86.27%
9 of 34
3%
LiveBench Instruction FollowingLiveBench
67.05%
23 of 34
3%
LiveBench CodingLiveBench
89.25%
2 of 34
3%
LiveBench Agentic CodingLiveBench
71.72%
2 of 34
3%
ArXivMathMathArena (ETH Zürich)
82.61%
4 of 7
3%
BrokenArXivMathArena (ETH Zürich)
77.3%
3 of 7
3%
FrontierMath Tiers 1–3Epoch AI
91.23%
4 of 38
2%
Mystery Game PuzzlesEpoch AI
71%
6 of 36
2%
LiveBench MathematicsLiveBench
97.08%
1 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
68.46%
12 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
57.87%
8 of 14
1%
GPQA DiamondEpoch AI
90.59%
28 of 35
1%
Chess PuzzlesEpoch AI
67%
5 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Fable 5.1
Place#3of 35 models with enough tests
Silicon score86.884.8 to 88.8 with other tests
Price for developers$10 / $50in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing92.0
Coding94.6
Math & reasoning92.1
AI agents82.4
Web research95.6
Understanding images99.6
Its tests
Claude Fable 5.1 took 19 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
70.8%
8 of 35
12%
Arena SearchArena (LMArena)
1237 rating
6 of 17
11%
Arena AgentArena (LMArena)
+14.1 points
6 of 39
9%
Arena WebDevArena (LMArena)
1743 rating
7 of 48
7%
MirrorCodeEpoch AI
93.33%
3 of 32
7%
Arena TextArena (LMArena)
1502 rating
7 of 41
6%
LiveBench ReasoningLiveBench
91.69%
4 of 34
5%
Arena VisionArena (LMArena)
1289 rating
10 of 30
5%
Arena DocumentArena (LMArena)
1515 rating
1 of 20
5%
Furniture AssemblyEpoch AI
70%
9 of 31
5%
LiveBench CodingLiveBench
86.38%
3 of 34
4%
LiveBench Agentic CodingLiveBench
66.06%
3 of 34
4%
EBR-benchEpoch AI
57.14%
7 of 21
4%
LiveBench LanguageLiveBench
89.5%
3 of 34
3%
LiveBench Instruction FollowingLiveBench
72.99%
11 of 34
3%
FrontierMath Tiers 1–3Epoch AI
90.18%
7 of 38
3%
Mystery Game PuzzlesEpoch AI
58%
10 of 36
3%
LiveBench MathematicsLiveBench
97.01%
2 of 34
2%
Chess PuzzlesEpoch AI
48%
17 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
GoogleClosed model
Gemini 4 Argon
Place#4of 35 models with enough tests
Silicon score85.483 to 87.8 with other tests
Price for developers$2 / $10in / out, per million tokens, until November 30. Then $4 / $20.
Each job on its own · scored in 4 of 6
Chat & writing98.5
Coding84.4
Math & reasoning90.1
AI agents79.7
Its tests
Gemini 4 Argon took 13 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
76%
3 of 35
18%
Arena AgentArena (LMArena)
+10 points
10 of 39
13%
Arena WebDevArena (LMArena)
1679 rating
14 of 48
11%
MirrorCodeEpoch AI
58.89%
13 of 32
11%
Arena TextArena (LMArena)
1522 rating
2 of 41
10%
Furniture AssemblyEpoch AI
72.5%
8 of 31
8%
EBR-benchEpoch AI
64.76%
6 of 21
7%
ArXivMathMathArena (ETH Zürich)
84.39%
3 of 7
6%
BrokenArXivMathArena (ETH Zürich)
47.89%
7 of 7
6%
FrontierMath Tiers 1–3Epoch AI
85.96%
13 of 38
4%
Mystery Game PuzzlesEpoch AI
64%
8 of 36
4%
GPQA DiamondEpoch AI
96.4%
2 of 35
2%
Chess PuzzlesEpoch AI
59%
9 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-6.1 Sol
Place#5of 35 models with enough tests
Silicon score85.383.5 to 87.1 with other tests
Price for developers$2 / $10in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing91.0
Coding85.5
Math & reasoning97.9
AI agents80.0
Web research95.5
Understanding images98.9
Its tests
GPT-6.1 Sol took 25 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
73.9%
5 of 35
10%
Arena SearchArena (LMArena)
1230 rating
8 of 17
9%
Arena VisionArena (LMArena)
1293 rating
8 of 30
9%
Arena AgentArena (LMArena)
+11.78 points
8 of 39
7%
Arena WebDevArena (LMArena)
1742 rating
8 of 48
6%
MirrorCodeEpoch AI
61.11%
11 of 32
6%
Arena TextArena (LMArena)
1481 rating
22 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
88.16%
3 of 13
4%
LiveBench ReasoningLiveBench
92.63%
2 of 34
4%
EBR-benchEpoch AI
54.29%
8 of 21
4%
Furniture AssemblyEpoch AI
80%
3 of 31
4%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
48.46%
3 of 13
4%
LiveBench LanguageLiveBench
90.13%
2 of 34
3%
LiveBench Instruction FollowingLiveBench
74.15%
9 of 34
3%
LiveBench CodingLiveBench
80.74%
14 of 34
3%
LiveBench Agentic CodingLiveBench
56.77%
16 of 34
3%
ArXivMathMathArena (ETH Zürich)
92.81%
2 of 7
3%
BrokenArXivMathArena (ETH Zürich)
83.09%
2 of 7
3%
FrontierMath Tiers 1–3Epoch AI
93.68%
2 of 38
2%
Mystery Game PuzzlesEpoch AI
80%
3 of 36
2%
LiveBench MathematicsLiveBench
96.83%
3 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
82.42%
2 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
82.11%
2 of 14
1%
GPQA DiamondEpoch AI
95.39%
5 of 35
1%
Chess PuzzlesEpoch AI
68%
4 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed modelFew tests yet
GPT-6.1 Terra
Place–None yet: few tests yet
Silicon score85.182.2 to 88 with other tests
Price for developers$1 / $5in / out, per million tokens
Each job on its own · scored in 2 of 6
Coding88.3
AI agents82.7
Its tests
GPT-6.1 Terra took 9 of our tests, from 2 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
57.3%
16 of 35
22%
Arena AgentArena (LMArena)
+11.34 points
9 of 39
17%
Arena WebDevArena (LMArena)
1681 rating
13 of 48
14%
MirrorCodeEpoch AI
76.67%
6 of 32
14%
Furniture AssemblyEpoch AI
79.17%
5 of 31
10%
EBR-benchEpoch AI
78.1%
3 of 21
8%
FrontierMath Tiers 1–3Epoch AI
90.88%
5 of 38
6%
Mystery Game PuzzlesEpoch AI
74%
4 of 36
6%
GPQA DiamondEpoch AI
94.32%
9 of 35
3%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it. Few tests yet: it has a job score of its own in only 2 of the 6 jobs, so the score may still move. It is listed, but gets no place.
OpenAIClosed model
GPT-6 Astra
Place#6of 35 models with enough tests
Silicon score84.983.1 to 86.7 with other tests
Price for developers$10 / $50in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing90.6
Coding86.0
Math & reasoning97.2
AI agents84.8
Web research96.9
Understanding images94.6
Its tests
GPT-6 Astra took 24 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
75.6%
4 of 35
10%
Arena SearchArena (LMArena)
1231 rating
7 of 17
9%
Arena AgentArena (LMArena)
+13.91 points
7 of 39
8%
Arena WebDevArena (LMArena)
1763 rating
5 of 48
6%
MirrorCodeEpoch AI
60%
12 of 32
6%
Arena TextArena (LMArena)
1474 rating
27 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
89.92%
1 of 13
5%
Arena VisionArena (LMArena)
1283 rating
15 of 30
5%
Arena DocumentArena (LMArena)
1467 rating
13 of 20
5%
Furniture AssemblyEpoch AI
80%
3 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
47.73%
5 of 13
5%
LiveBench ReasoningLiveBench
92.65%
1 of 34
4%
EBR-benchEpoch AI
76.19%
4 of 21
4%
LiveBench LanguageLiveBench
89.43%
4 of 34
3%
LiveBench Instruction FollowingLiveBench
75.58%
7 of 34
3%
LiveBench CodingLiveBench
80.36%
16 of 34
3%
LiveBench Agentic CodingLiveBench
57.32%
13 of 34
3%
FrontierMath Tiers 1–3Epoch AI
93.68%
2 of 38
2%
Mystery Game PuzzlesEpoch AI
84%
2 of 36
2%
LiveBench MathematicsLiveBench
96.81%
4 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
81.98%
3 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
83.08%
1 of 14
1%
GPQA DiamondEpoch AI
95.77%
3 of 35
1%
Chess PuzzlesEpoch AI
74%
3 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Opus 5
Place#7of 35 models with enough tests
Silicon score82.580.6 to 84.4 with other tests
Price for developers$5 / $25in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing85.3
Coding89.8
Math & reasoning90.2
AI agents74.9
Web research93.7
Understanding images96.7
Its tests
Claude Opus 5 took 21 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
59.9%
15 of 35
11%
Arena SearchArena (LMArena)
1239 rating
5 of 17
10%
Arena AgentArena (LMArena)
+6.18 points
15 of 39
8%
τ³-bench Bankingτ³-bench (Sierra)
48.71%
2 of 12
8%
Arena WebDevArena (LMArena)
1691 rating
11 of 48
7%
MirrorCodeEpoch AI
87.78%
4 of 32
7%
Arena TextArena (LMArena)
1490 rating
12 of 41
6%
Arena VisionArena (LMArena)
1285 rating
13 of 30
5%
Arena DocumentArena (LMArena)
1509 rating
2 of 20
5%
Furniture AssemblyEpoch AI
60.83%
10 of 31
5%
LiveBench ReasoningLiveBench
91.21%
7 of 34
4%
EBR-benchEpoch AI
45.71%
10 of 21
4%
LiveBench LanguageLiveBench
88.69%
5 of 34
3%
LiveBench Instruction FollowingLiveBench
63.77%
29 of 34
3%
LiveBench CodingLiveBench
81.45%
11 of 34
3%
LiveBench Agentic CodingLiveBench
65.2%
4 of 34
3%
FrontierMath Tiers 1–3Epoch AI
85.61%
15 of 38
2%
Mystery Game PuzzlesEpoch AI
59%
9 of 36
2%
LiveBench MathematicsLiveBench
95.73%
11 of 34
2%
GPQA DiamondEpoch AI
93.88%
12 of 35
1%
Chess PuzzlesEpoch AI
43%
24 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Sonnet 5.5
Place#8of 35 models with enough tests
Silicon score82.380.1 to 84.5 with other tests
Price for developers$2 / $10in / out, per million tokens
Each job on its own · scored in 4 of 6
Chat & writing80.4
Coding95.8
Math & reasoning93.3
Understanding images87.4
Its tests
Claude Sonnet 5.5 took 17 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
46.5%
26 of 35
14%
Arena VisionArena (LMArena)
1263 rating
23 of 30
13%
Arena AgentArena (LMArena)
+15.29 points
4 of 39
10%
Arena WebDevArena (LMArena)
1772 rating
3 of 48
8%
MirrorCodeEpoch AI
97.78%
1 of 32
8%
Arena TextArena (LMArena)
1473 rating
28 of 41
7%
LiveBench ReasoningLiveBench
91.63%
6 of 34
6%
Furniture AssemblyEpoch AI
75%
7 of 31
6%
LiveBench LanguageLiveBench
83.43%
15 of 34
4%
LiveBench Instruction FollowingLiveBench
70.52%
18 of 34
4%
LiveBench CodingLiveBench
91.37%
1 of 34
4%
LiveBench Agentic CodingLiveBench
56.31%
19 of 34
4%
FrontierMath Tiers 1–3Epoch AI
88.77%
10 of 38
3%
Mystery Game PuzzlesEpoch AI
65%
7 of 36
3%
LiveBench MathematicsLiveBench
96.69%
5 of 34
3%
GPQA DiamondEpoch AI
95.58%
4 of 35
1%
Chess PuzzlesEpoch AI
59%
9 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Fable 5
Place#9of 35 models with enough tests
Silicon score81.379.4 to 83.2 with other tests
Price for developers$10 / $50in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing93.3
Coding85.8
Math & reasoning85.9
AI agents70.0
Web research93.3
Understanding images95.5
Its tests
Claude Fable 5 took 21 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
70.7%
9 of 35
11%
Arena SearchArena (LMArena)
1228 rating
10 of 17
10%
Arena AgentArena (LMArena)
+6.22 points
14 of 39
8%
τ³-bench Bankingτ³-bench (Sierra)
39.69%
7 of 12
8%
Arena WebDevArena (LMArena)
1626 rating
24 of 48
7%
MirrorCodeEpoch AI
70%
7 of 32
7%
Arena TextArena (LMArena)
1504 rating
5 of 41
6%
Arena VisionArena (LMArena)
1309 rating
1 of 30
5%
Arena DocumentArena (LMArena)
1497 rating
4 of 20
5%
Furniture AssemblyEpoch AI
35.83%
20 of 31
5%
LiveBench ReasoningLiveBench
89.65%
11 of 34
4%
EBR-benchEpoch AI
39.52%
13 of 21
4%
LiveBench LanguageLiveBench
90.68%
1 of 34
3%
LiveBench Instruction FollowingLiveBench
75.77%
6 of 34
3%
LiveBench CodingLiveBench
85.99%
4 of 34
3%
LiveBench Agentic CodingLiveBench
62.17%
7 of 34
3%
FrontierMath Tiers 1–3Epoch AI
87.02%
12 of 38
2%
Mystery Game PuzzlesEpoch AI
52%
14 of 36
2%
LiveBench MathematicsLiveBench
95.99%
8 of 34
2%
GPQA DiamondEpoch AI
85.86%
35 of 35
1%
Chess PuzzlesEpoch AI
47%
19 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-5.6 Sol
Place#10of 35 models with enough tests
Silicon score80.678.8 to 82.4 with other tests
Price for developers$4 / $20in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing87.4
Coding75.9
Math & reasoning91.9
AI agents72.6
Web research100.0
Understanding images92.6
Its tests
GPT-5.6 Sol took 23 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
69.7%
10 of 35
10%
Arena SearchArena (LMArena)
1258 rating
2 of 17
9%
Arena AgentArena (LMArena)
+3.72 points
18 of 39
7%
τ³-bench Bankingτ³-bench (Sierra)
46.91%
4 of 12
7%
Arena WebDevArena (LMArena)
1616 rating
29 of 48
6%
MirrorCodeEpoch AI
20%
31 of 32
6%
Arena TextArena (LMArena)
1485 rating
16 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
88.48%
2 of 13
5%
Arena VisionArena (LMArena)
1288 rating
11 of 30
5%
Arena DocumentArena (LMArena)
1480 rating
8 of 20
5%
Furniture AssemblyEpoch AI
56.67%
12 of 31
5%
LiveBench ReasoningLiveBench
91.65%
5 of 34
4%
EBR-benchEpoch AI
44.76%
11 of 21
4%
LiveBench LanguageLiveBench
87.68%
7 of 34
3%
LiveBench Instruction FollowingLiveBench
71.85%
14 of 34
3%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
68.75%
11 of 15
3%
LiveBench CodingLiveBench
83.94%
5 of 34
3%
LiveBench Agentic CodingLiveBench
56.21%
20 of 34
3%
FrontierMath Tiers 1–3Epoch AI
89.12%
9 of 38
2%
Mystery Game PuzzlesEpoch AI
58%
10 of 36
2%
LiveBench MathematicsLiveBench
96.2%
7 of 34
2%
GPQA DiamondEpoch AI
93.5%
14 of 35
1%
Chess PuzzlesEpoch AI
59%
9 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Alibaba (Qwen)Closed model
Qwen3.8 Max
Place#11of 35 models with enough tests
Silicon score78.776.7 to 80.7 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 6 of 6
Chat & writing82.0
Coding80.7
Math & reasoning82.5
AI agents74.4
Web research96.0
Understanding images92.0
Its tests
Qwen3.8 Max took 19 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
47.3%
24 of 35
11%
Arena SearchArena (LMArena)
1260 rating
1 of 17
10%
Arena VisionArena (LMArena)
1299 rating
3 of 30
10%
Arena AgentArena (LMArena)
-0.87 points
33 of 39
8%
τ³-bench Bankingτ³-bench (Sierra)
55.15%
1 of 12
8%
Arena WebDevArena (LMArena)
1676 rating
15 of 48
7%
MirrorCodeEpoch AI
52.22%
17 of 32
7%
Arena TextArena (LMArena)
1484 rating
18 of 41
6%
LiveBench ReasoningLiveBench
88.21%
19 of 34
5%
Furniture AssemblyEpoch AI
20%
31 of 31
5%
LiveBench LanguageLiveBench
79.69%
24 of 34
3%
LiveBench Instruction FollowingLiveBench
74.08%
10 of 34
3%
LiveBench CodingLiveBench
72.87%
31 of 34
3%
LiveBench Agentic CodingLiveBench
64.65%
5 of 34
3%
FrontierMath Tiers 1–3Epoch AI
74.74%
21 of 38
2%
Mystery Game PuzzlesEpoch AI
38%
18 of 36
2%
LiveBench MathematicsLiveBench
91.31%
23 of 34
2%
GPQA DiamondEpoch AI
92.68%
20 of 35
1%
Chess PuzzlesEpoch AI
46%
21 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Moonshot AI (Kimi)Open modelFew tests yet
Kimi K3.1 Pro
Place–None yet: few tests yet
Silicon score78.775.1 to 82.3 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 0 of 6
It does not have enough tests in any single job for a job score of its own yet.
Its tests
Kimi K3.1 Pro took 5 of our tests, from 2 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
Arena AgentArena (LMArena)
+3.95 points
17 of 39
30%
Arena WebDevArena (LMArena)
1686 rating
12 of 48
25%
Arena TextArena (LMArena)
1485 rating
17 of 41
21%
ArXivMathMathArena (ETH Zürich)
58.27%
5 of 7
12%
BrokenArXivMathArena (ETH Zürich)
61.94%
5 of 7
12%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it. Few tests yet: it has a job score of its own in only 0 of the 6 jobs, so the score may still move. It is listed, but gets no place.
MetaClosed model
Muse Spark 1.3
Place#12of 35 models with enough tests
Silicon score78.376.2 to 80.4 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 3 of 6
Chat & writing89.5
Coding83.2
Math & reasoning83.6
Its tests
Muse Spark 1.3 took 15 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
Arena AgentArena (LMArena)
+1.2 points
24 of 39
13%
Arena WebDevArena (LMArena)
1655 rating
19 of 48
11%
MirrorCodeEpoch AI
54.44%
14 of 32
11%
Arena TextArena (LMArena)
1497 rating
10 of 41
9%
Arena VisionArena (LMArena)
1292 rating
9 of 30
8%
Arena DocumentArena (LMArena)
1473 rating
11 of 20
8%
LiveBench ReasoningLiveBench
89.65%
11 of 34
7%
LiveBench LanguageLiveBench
82.79%
19 of 34
5%
LiveBench Instruction FollowingLiveBench
78%
4 of 34
5%
LiveBench CodingLiveBench
81.06%
13 of 34
5%
LiveBench Agentic CodingLiveBench
64.09%
6 of 34
5%
FrontierMath Tiers 1–3Epoch AI
74.39%
22 of 38
4%
Mystery Game PuzzlesEpoch AI
25%
30 of 36
4%
LiveBench MathematicsLiveBench
95.95%
9 of 34
4%
Chess PuzzlesEpoch AI
37%
28 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-5.5
Place#13of 35 models with enough tests
Silicon score77.876 to 79.6 with other tests
Price for developers$5 / $30in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing87.0
Coding71.3
Math & reasoning88.5
AI agents70.7
Web research96.0
Understanding images91.5
Its tests
GPT-5.5 took 23 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
63%
13 of 35
11%
Arena SearchArena (LMArena)
1246 rating
4 of 17
10%
τ³-bench Bankingτ³-bench (Sierra)
44.59%
5 of 12
8%
Arena WebDevArena (LMArena)
1512 rating
45 of 48
7%
MirrorCodeEpoch AI
23.33%
29 of 32
7%
Arena TextArena (LMArena)
1481 rating
21 of 41
6%
Arena VisionArena (LMArena)
1285 rating
14 of 30
5%
Arena DocumentArena (LMArena)
1484 rating
6 of 20
5%
Furniture AssemblyEpoch AI
44.17%
15 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
45.77%
6 of 13
5%
LiveBench ReasoningLiveBench
89.65%
11 of 34
4%
EBR-benchEpoch AI
34.29%
14 of 21
4%
LiveBench LanguageLiveBench
87.36%
8 of 34
3%
LiveBench Instruction FollowingLiveBench
70.73%
17 of 34
3%
LiveBench CodingLiveBench
82.15%
7 of 34
3%
LiveBench Agentic CodingLiveBench
53.99%
24 of 34
3%
FrontierMath Tiers 1–3Epoch AI
85.26%
16 of 38
2%
Mystery Game PuzzlesEpoch AI
56%
12 of 36
2%
LiveBench MathematicsLiveBench
95.86%
10 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
76.19%
5 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
78.04%
5 of 14
1%
GPQA DiamondEpoch AI
90.66%
27 of 35
1%
Chess PuzzlesEpoch AI
63%
7 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
GoogleClosed model
Gemini 3.7 Flash
Place#14of 35 models with enough tests
Silicon score77.175.2 to 79 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 5 of 6
Chat & writing90.6
Coding76.6
Math & reasoning84.3
Web research87.7
Understanding images91.7
Its tests
Gemini 3.7 Flash took 22 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
69.2%
12 of 35
12%
Arena AgentArena (LMArena)
-3.06 points
36 of 39
9%
Arena WebDevArena (LMArena)
1588 rating
34 of 48
7%
MirrorCodeEpoch AI
41.11%
20 of 32
7%
Arena TextArena (LMArena)
1489 rating
13 of 41
6%
FACTS SearchFACTS (Google DeepMind and Kaggle)
80.89%
9 of 13
5%
LiveBench ReasoningLiveBench
87.8%
20 of 34
5%
Arena VisionArena (LMArena)
1294 rating
6 of 30
5%
Arena DocumentArena (LMArena)
1476 rating
9 of 20
5%
Furniture AssemblyEpoch AI
26.67%
26 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
49.83%
2 of 13
5%
LiveBench CodingLiveBench
78.89%
22 of 34
4%
LiveBench Agentic CodingLiveBench
58.28%
12 of 34
4%
LiveBench LanguageLiveBench
85.46%
11 of 34
3%
LiveBench Instruction FollowingLiveBench
79.93%
2 of 34
3%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
75.54%
6 of 15
2%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
78.72%
4 of 14
2%
FrontierMath Tiers 1–3Epoch AI
71.58%
24 of 38
2%
Mystery Game PuzzlesEpoch AI
37%
19 of 36
2%
LiveBench MathematicsLiveBench
93.47%
16 of 34
2%
GPQA DiamondEpoch AI
94.82%
7 of 35
1%
Chess PuzzlesEpoch AI
66%
6 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
GoogleClosed model
Gemini 3.8 Flash
Place#15of 35 models with enough tests
Silicon score76.975 to 78.8 with other tests
Price for developers$0.75 / $3.75in / out, per million tokens, until December 31. Then $1.50 / $7.50.
Each job on its own · scored in 5 of 6
Chat & writing92.2
Coding72.7
Math & reasoning85.0
Web research82.3
Understanding images93.1
Its tests
Gemini 3.8 Flash took 22 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
69.7%
10 of 35
12%
Arena AgentArena (LMArena)
+1.44 points
23 of 39
9%
Arena WebDevArena (LMArena)
1583 rating
35 of 48
7%
MirrorCodeEpoch AI
38.89%
23 of 32
7%
Arena TextArena (LMArena)
1497 rating
9 of 41
6%
FACTS SearchFACTS (Google DeepMind and Kaggle)
76.33%
13 of 13
5%
LiveBench ReasoningLiveBench
89.29%
14 of 34
5%
Arena VisionArena (LMArena)
1293 rating
7 of 30
5%
Arena DocumentArena (LMArena)
1482 rating
7 of 20
5%
Furniture AssemblyEpoch AI
31.67%
24 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
50.7%
1 of 13
5%
LiveBench CodingLiveBench
72.49%
33 of 34
4%
LiveBench Agentic CodingLiveBench
54.24%
23 of 34
4%
LiveBench LanguageLiveBench
87.79%
6 of 34
3%
LiveBench Instruction FollowingLiveBench
81.41%
1 of 34
3%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
73.09%
8 of 15
2%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
77.34%
6 of 14
2%
FrontierMath Tiers 1–3Epoch AI
68.42%
28 of 38
2%
Mystery Game PuzzlesEpoch AI
47%
16 of 36
2%
LiveBench MathematicsLiveBench
91.56%
22 of 34
2%
GPQA DiamondEpoch AI
95.39%
5 of 35
1%
Chess PuzzlesEpoch AI
58%
12 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-6 Sol
Place#16of 35 models with enough tests
Silicon score75.773.8 to 77.6 with other tests
Price for developers$2 / $10in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing83.4
Coding80.6
Math & reasoning89.4
AI agents74.8
Web research82.0
Understanding images84.7
Its tests
GPT-6 Sol took 23 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
60.7%
14 of 35
10%
Arena SearchArena (LMArena)
1174 rating
16 of 17
9%
Arena VisionArena (LMArena)
1258 rating
25 of 30
9%
Arena AgentArena (LMArena)
+6.25 points
13 of 39
8%
Arena WebDevArena (LMArena)
1673 rating
16 of 48
6%
MirrorCodeEpoch AI
48.89%
18 of 32
6%
Arena TextArena (LMArena)
1457 rating
35 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
86.15%
5 of 13
5%
Furniture AssemblyEpoch AI
58.33%
11 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
43.06%
7 of 13
5%
LiveBench ReasoningLiveBench
88.65%
18 of 34
4%
EBR-benchEpoch AI
53.33%
9 of 21
4%
LiveBench LanguageLiveBench
85.3%
13 of 34
3%
LiveBench Instruction FollowingLiveBench
68.57%
21 of 34
3%
LiveBench CodingLiveBench
81.77%
10 of 34
3%
LiveBench Agentic CodingLiveBench
52.88%
26 of 34
3%
FrontierMath Tiers 1–3Epoch AI
89.82%
8 of 38
2%
Mystery Game PuzzlesEpoch AI
56%
12 of 36
2%
LiveBench MathematicsLiveBench
96.36%
6 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
85.22%
1 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
75.59%
7 of 14
1%
GPQA DiamondEpoch AI
94.32%
9 of 35
1%
Chess PuzzlesEpoch AI
56%
13 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Moonshot AI (Kimi)Open model
Kimi K3
Place#17of 35 models with enough tests
Silicon score74.872.7 to 76.9 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 4 of 6
Chat & writing84.1
Coding83.0
Math & reasoning80.2
AI agents63.8
Its tests
Kimi K3 took 17 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
50.6%
21 of 35
14%
Arena AgentArena (LMArena)
-0.24 points
30 of 39
10%
τ³-bench Bankingτ³-bench (Sierra)
37.11%
9 of 12
10%
Arena WebDevArena (LMArena)
1658 rating
18 of 48
9%
MirrorCodeEpoch AI
54.44%
14 of 32
9%
Arena TextArena (LMArena)
1487 rating
14 of 41
7%
Furniture AssemblyEpoch AI
34.17%
21 of 31
7%
LiveBench ReasoningLiveBench
90.67%
8 of 34
6%
LiveBench LanguageLiveBench
85.53%
10 of 34
4%
LiveBench Instruction FollowingLiveBench
71.36%
16 of 34
4%
LiveBench CodingLiveBench
81.45%
11 of 34
4%
LiveBench Agentic CodingLiveBench
62.17%
7 of 34
4%
FrontierMath Tiers 1–3Epoch AI
72.18%
23 of 38
3%
Mystery Game PuzzlesEpoch AI
26%
29 of 36
3%
LiveBench MathematicsLiveBench
84.44%
32 of 34
3%
GPQA DiamondEpoch AI
93.12%
18 of 35
1%
Chess PuzzlesEpoch AI
37%
28 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
DeepSeekOpen model
DeepSeek V4.1 Flash
Place#18of 35 models with enough tests
Silicon score74.472.4 to 76.4 with other tests
Price for developers$0.30 / $1.20in / out, per million tokens
Each job on its own · scored in 4 of 6
Chat & writing79.1
Coding87.7
Math & reasoning81.1
Understanding images81.2
Its tests
DeepSeek V4.1 Flash took 19 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
42.9%
28 of 35
13%
Arena VisionArena (LMArena)
1263 rating
22 of 30
12%
Arena AgentArena (LMArena)
+1.13 points
26 of 39
9%
Arena WebDevArena (LMArena)
1661 rating
17 of 48
8%
MirrorCodeEpoch AI
64.44%
9 of 32
8%
Arena TextArena (LMArena)
1475 rating
25 of 41
7%
Furniture AssemblyEpoch AI
34.17%
21 of 31
6%
LiveBench ReasoningLiveBench
86.69%
24 of 34
5%
LiveBench CodingLiveBench
80.04%
17 of 34
4%
LiveBench Agentic CodingLiveBench
77.27%
1 of 34
4%
ArXivMathMathArena (ETH Zürich)
42.48%
6 of 7
4%
BrokenArXivMathArena (ETH Zürich)
48.7%
6 of 7
4%
LiveBench LanguageLiveBench
81.19%
21 of 34
3%
LiveBench Instruction FollowingLiveBench
70.03%
19 of 34
3%
FrontierMath Tiers 1–3Epoch AI
71.23%
25 of 38
3%
Mystery Game PuzzlesEpoch AI
49%
15 of 36
3%
LiveBench MathematicsLiveBench
93.29%
17 of 34
3%
GPQA DiamondEpoch AI
93.18%
17 of 35
1%
Chess PuzzlesEpoch AI
48%
17 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Opus 4.7
Place#19of 35 models with enough tests
Silicon score74.272.4 to 76 with other tests
Price for developers$5 / $25in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing81.4
Coding75.1
Math & reasoning79.6
AI agents65.2
Web research89.7
Understanding images89.2
Its tests
Claude Opus 4.7 took 23 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
51.7%
20 of 35
11%
Arena SearchArena (LMArena)
1230 rating
9 of 17
10%
τ³-bench Bankingτ³-bench (Sierra)
40.21%
6 of 12
8%
Arena WebDevArena (LMArena)
1557 rating
38 of 48
7%
MirrorCodeEpoch AI
43.33%
19 of 32
7%
Arena TextArena (LMArena)
1501 rating
8 of 41
6%
Arena VisionArena (LMArena)
1300 rating
2 of 30
5%
Arena DocumentArena (LMArena)
1496 rating
5 of 20
5%
Furniture AssemblyEpoch AI
33.33%
23 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
35.84%
11 of 13
5%
LiveBench ReasoningLiveBench
87.19%
22 of 34
4%
EBR-benchEpoch AI
19.05%
17 of 21
4%
LiveBench LanguageLiveBench
77.91%
26 of 34
3%
LiveBench Instruction FollowingLiveBench
66.74%
24 of 34
3%
LiveBench CodingLiveBench
82.09%
8 of 34
3%
LiveBench Agentic CodingLiveBench
50.66%
29 of 34
3%
FrontierMath Tiers 1–3Epoch AI
70.18%
26 of 38
2%
Mystery Game PuzzlesEpoch AI
28%
28 of 36
2%
LiveBench MathematicsLiveBench
92.85%
19 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
61.73%
14 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
38.82%
12 of 14
1%
GPQA DiamondEpoch AI
90.15%
32 of 35
1%
Chess PuzzlesEpoch AI
41%
26 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Opus 4.8
Place#20of 35 models with enough tests
Silicon score74.072.2 to 75.8 with other tests
Price for developers$5 / $25in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing81.1
Coding78.1
Math & reasoning84.8
AI agents66.0
Web research81.8
Understanding images87.9
Its tests
Claude Opus 4.8 took 24 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
53%
18 of 35
10%
Arena SearchArena (LMArena)
1202 rating
14 of 17
9%
τ³-bench Bankingτ³-bench (Sierra)
39.69%
7 of 12
8%
Arena WebDevArena (LMArena)
1555 rating
39 of 48
6%
MirrorCodeEpoch AI
65.56%
8 of 32
6%
Arena TextArena (LMArena)
1482 rating
19 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
79.6%
10 of 13
5%
Arena VisionArena (LMArena)
1286 rating
12 of 30
5%
Arena DocumentArena (LMArena)
1475 rating
10 of 20
5%
Furniture AssemblyEpoch AI
42.5%
17 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
39.77%
9 of 13
5%
LiveBench ReasoningLiveBench
89.19%
15 of 34
4%
EBR-benchEpoch AI
28.57%
16 of 21
4%
LiveBench LanguageLiveBench
79.66%
25 of 34
3%
LiveBench Instruction FollowingLiveBench
72.03%
12 of 34
3%
LiveBench CodingLiveBench
81.83%
9 of 34
3%
LiveBench Agentic CodingLiveBench
50.51%
30 of 34
3%
FrontierMath Tiers 1–3Epoch AI
80%
18 of 38
2%
Mystery Game PuzzlesEpoch AI
36%
20 of 36
2%
LiveBench MathematicsLiveBench
94.32%
15 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
69.68%
10 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
25.06%
14 of 14
1%
GPQA DiamondEpoch AI
91.04%
24 of 35
1%
Chess PuzzlesEpoch AI
52%
15 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Opus 4.6
Place#21of 35 models with enough tests
Silicon score72.570.7 to 74.3 with other tests
Price for developers$5 / $25in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing80.5
Coding71.4
Math & reasoning77.7
AI agents53.8
Web research94.0
Understanding images90.0
Its tests
Claude Opus 4.6 took 23 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
47%
25 of 35
11%
Arena SearchArena (LMArena)
1252 rating
3 of 17
10%
τ³-bench Bankingτ³-bench (Sierra)
27.32%
11 of 12
8%
Arena WebDevArena (LMArena)
1547 rating
41 of 48
7%
MirrorCodeEpoch AI
33.33%
25 of 32
7%
Arena TextArena (LMArena)
1503 rating
6 of 41
6%
Arena VisionArena (LMArena)
1299 rating
4 of 30
5%
Arena DocumentArena (LMArena)
1508 rating
3 of 20
5%
Furniture AssemblyEpoch AI
28.33%
25 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
36.75%
10 of 13
5%
LiveBench ReasoningLiveBench
88.67%
17 of 34
4%
EBR-benchEpoch AI
12.7%
19 of 21
4%
LiveBench LanguageLiveBench
83.27%
16 of 34
3%
LiveBench Instruction FollowingLiveBench
63.31%
30 of 34
3%
LiveBench CodingLiveBench
78.18%
25 of 34
3%
LiveBench Agentic CodingLiveBench
48.99%
31 of 34
3%
FrontierMath Tiers 1–3Epoch AI
65.96%
29 of 38
2%
Mystery Game PuzzlesEpoch AI
25%
30 of 36
2%
LiveBench MathematicsLiveBench
89.32%
27 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
54.91%
15 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
32.95%
13 of 14
1%
GPQA DiamondEpoch AI
90.53%
29 of 35
1%
Chess PuzzlesEpoch AI
21%
35 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-5.6 Terra
Place#22of 35 models with enough tests
Silicon score72.070.1 to 73.9 with other tests
Price for developers$2 / $12in / out, per million tokens
Each job on its own · scored in 5 of 6
Chat & writing78.0
Coding73.0
Math & reasoning87.5
Web research80.4
Understanding images86.7
Its tests
GPT-5.6 Terra took 21 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
43.2%
27 of 35
12%
Arena AgentArena (LMArena)
-2.72 points
35 of 39
9%
Arena WebDevArena (LMArena)
1520 rating
43 of 48
8%
MirrorCodeEpoch AI
41.11%
20 of 32
8%
Arena TextArena (LMArena)
1468 rating
30 of 41
7%
FACTS SearchFACTS (Google DeepMind and Kaggle)
81.74%
7 of 13
6%
Arena VisionArena (LMArena)
1266 rating
19 of 30
6%
Arena DocumentArena (LMArena)
1468 rating
12 of 20
6%
Furniture AssemblyEpoch AI
54.17%
13 of 31
6%
LiveBench ReasoningLiveBench
90.63%
9 of 34
5%
LiveBench CodingLiveBench
78.25%
24 of 34
4%
LiveBench Agentic CodingLiveBench
54.95%
21 of 34
4%
LiveBench LanguageLiveBench
82.89%
17 of 34
3%
LiveBench Instruction FollowingLiveBench
64.62%
26 of 34
3%
FrontierMath Tiers 1–3Epoch AI
85.96%
13 of 38
3%
Mystery Game PuzzlesEpoch AI
35%
21 of 36
3%
LiveBench MathematicsLiveBench
94.91%
14 of 34
3%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
71.21%
9 of 15
2%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
55.04%
10 of 14
2%
GPQA DiamondEpoch AI
93.31%
16 of 35
1%
Chess PuzzlesEpoch AI
47%
19 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
xAIClosed model
Grok 4.6
Place#23of 35 models with enough tests
Silicon score71.869.8 to 73.8 with other tests
Price for developers$2 / $6in / out, per million tokens
Each job on its own · scored in 5 of 6
Chat & writing79.1
Coding76.6
Math & reasoning83.2
AI agents65.3
Understanding images80.0
Its tests
Grok 4.6 took 19 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
49.3%
22 of 35
13%
Arena AgentArena (LMArena)
-0.92 points
34 of 39
10%
Arena WebDevArena (LMArena)
1618 rating
28 of 48
8%
MirrorCodeEpoch AI
41.11%
20 of 32
8%
Arena TextArena (LMArena)
1457 rating
36 of 41
7%
Arena VisionArena (LMArena)
1263 rating
21 of 30
6%
Arena DocumentArena (LMArena)
1438 rating
19 of 20
6%
Furniture AssemblyEpoch AI
40%
19 of 31
6%
LiveBench ReasoningLiveBench
90.51%
10 of 34
5%
EBR-benchEpoch AI
30.48%
15 of 21
5%
LiveBench CodingLiveBench
76.78%
29 of 34
4%
LiveBench Agentic CodingLiveBench
57.02%
15 of 34
4%
LiveBench LanguageLiveBench
83.7%
14 of 34
3%
LiveBench Instruction FollowingLiveBench
71.87%
13 of 34
3%
FrontierMath Tiers 1–3Epoch AI
65.96%
29 of 38
3%
Mystery Game PuzzlesEpoch AI
34%
23 of 36
3%
LiveBench MathematicsLiveBench
92.57%
21 of 34
3%
GPQA DiamondEpoch AI
94%
11 of 35
1%
Chess PuzzlesEpoch AI
33%
33 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
DeepSeekOpen model
DeepSeek V4 Pro 0813
Place#24of 35 models with enough tests
Silicon score71.869.4 to 74.2 with other tests
Price for developers$1.32 / $3.96in / out, per million tokens
Each job on its own · scored in 3 of 6
Chat & writing79.3
Coding73.7
Math & reasoning80.6
Its tests
DeepSeek V4 Pro 0813 took 14 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
52.91%
19 of 35
19%
Arena WebDevArena (LMArena)
1581 rating
36 of 48
12%
MirrorCodeEpoch AI
33.33%
25 of 32
12%
Arena TextArena (LMArena)
1464 rating
32 of 41
10%
LiveBench ReasoningLiveBench
85.84%
25 of 34
8%
LiveBench CodingLiveBench
77.16%
27 of 34
6%
LiveBench Agentic CodingLiveBench
54.95%
21 of 34
6%
LiveBench LanguageLiveBench
82.07%
20 of 34
5%
LiveBench Instruction FollowingLiveBench
67.7%
22 of 34
5%
FrontierMath Tiers 1–3Epoch AI
64.56%
32 of 38
4%
Mystery Game PuzzlesEpoch AI
43%
17 of 36
4%
LiveBench MathematicsLiveBench
95.09%
13 of 34
4%
GPQA DiamondEpoch AI
91.67%
22 of 35
2%
Chess PuzzlesEpoch AI
46%
21 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Z.ai (GLM)Open model
GLM-5.3
Place#25of 35 models with enough tests
Silicon score71.869.4 to 74.2 with other tests
Price for developers$1.40 / $4.40in / out, per million tokens
Each job on its own · scored in 3 of 6
Chat & writing79.1
Coding80.2
Math & reasoning76.7
Its tests
GLM-5.3 took 14 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
41%
30 of 35
19%
Arena AgentArena (LMArena)
+0.03 points
29 of 39
14%
Arena WebDevArena (LMArena)
1622 rating
27 of 48
12%
Arena TextArena (LMArena)
1482 rating
20 of 41
10%
LiveBench ReasoningLiveBench
85.8%
26 of 34
8%
LiveBench CodingLiveBench
78.95%
19 of 34
6%
LiveBench Agentic CodingLiveBench
60.91%
10 of 34
6%
LiveBench LanguageLiveBench
79.86%
23 of 34
5%
LiveBench Instruction FollowingLiveBench
69.3%
20 of 34
5%
FrontierMath Tiers 1–3Epoch AI
68.77%
27 of 38
4%
Mystery Game PuzzlesEpoch AI
33%
25 of 36
4%
LiveBench MathematicsLiveBench
87.9%
29 of 34
4%
GPQA DiamondEpoch AI
90.91%
25 of 35
2%
Chess PuzzlesEpoch AI
37%
28 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
xAIClosed model
Grok 4.5
Place#26of 35 models with enough tests
Silicon score71.369.4 to 73.2 with other tests
Price for developers$2 / $6in / out, per million tokens
Each job on its own · scored in 5 of 6
Chat & writing80.2
Coding69.7
Math & reasoning77.6
Web research83.4
Understanding images82.1
Its tests
Grok 4.5 took 21 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
48.3%
23 of 35
11%
Arena SearchArena (LMArena)
1216 rating
11 of 17
10%
τ³-bench Bankingτ³-bench (Sierra)
47.94%
3 of 12
8%
Arena WebDevArena (LMArena)
1552 rating
40 of 48
7%
MirrorCodeEpoch AI
32.22%
27 of 32
7%
Arena TextArena (LMArena)
1466 rating
31 of 41
6%
FACTS SearchFACTS (Google DeepMind and Kaggle)
79.44%
11 of 13
5%
LiveBench ReasoningLiveBench
87.17%
23 of 34
5%
Arena VisionArena (LMArena)
1280 rating
16 of 30
5%
Arena DocumentArena (LMArena)
1451 rating
16 of 20
5%
Furniture AssemblyEpoch AI
22.5%
28 of 31
5%
LiveBench LanguageLiveBench
82.8%
18 of 34
3%
LiveBench Instruction FollowingLiveBench
71.53%
15 of 34
3%
LiveBench CodingLiveBench
68.59%
34 of 34
3%
LiveBench Agentic CodingLiveBench
56.46%
18 of 34
3%
FrontierMath Tiers 1–3Epoch AI
57.19%
35 of 38
3%
GPQA DiamondEpoch AI
93.43%
15 of 35
2%
Chess PuzzlesEpoch AI
35%
32 of 37
2%
LiveBench MathematicsLiveBench
90.82%
25 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
74.52%
7 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
46.83%
11 of 14
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
xAIClosed model
Grok 4.7
Place#27of 35 models with enough tests
Silicon score70.568.3 to 72.7 with other tests
Price for developers$2 / $6in / out, per million tokens
Each job on its own · scored in 4 of 6
Chat & writing78.5
Coding78.7
Math & reasoning75.6
Understanding images74.1
Its tests
Grok 4.7 took 17 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
56%
17 of 35
14%
Arena VisionArena (LMArena)
1249 rating
28 of 30
13%
Arena AgentArena (LMArena)
+1.78 points
20 of 39
10%
Arena WebDevArena (LMArena)
1640 rating
22 of 48
8%
MirrorCodeEpoch AI
54.44%
14 of 32
8%
Arena TextArena (LMArena)
1443 rating
38 of 41
7%
LiveBench ReasoningLiveBench
82.65%
30 of 34
6%
Furniture AssemblyEpoch AI
20.83%
30 of 31
6%
LiveBench LanguageLiveBench
80.14%
22 of 34
4%
LiveBench Instruction FollowingLiveBench
75.28%
8 of 34
4%
LiveBench CodingLiveBench
77.16%
27 of 34
4%
LiveBench Agentic CodingLiveBench
53.99%
24 of 34
4%
FrontierMath Tiers 1–3Epoch AI
52.98%
38 of 38
3%
Mystery Game PuzzlesEpoch AI
29%
27 of 36
3%
LiveBench MathematicsLiveBench
95.68%
12 of 34
3%
GPQA DiamondEpoch AI
92.74%
19 of 35
1%
Chess PuzzlesEpoch AI
41%
26 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Alibaba (Qwen)Open modelFew tests yet
Qwen3.8 Flash Next
Place–None yet: few tests yet
Silicon score70.267.3 to 73.1 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 1 of 6
Coding77.1
Its tests
Qwen3.8 Flash Next took 8 of our tests, from 2 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
Arena AgentArena (LMArena)
-5.18 points
38 of 39
24%
Arena WebDevArena (LMArena)
1625 rating
25 of 48
20%
LiveBench ReasoningLiveBench
87.38%
21 of 34
13%
LiveBench CodingLiveBench
72.55%
32 of 34
10%
LiveBench Agentic CodingLiveBench
61.62%
9 of 34
10%
LiveBench LanguageLiveBench
74.64%
31 of 34
8%
LiveBench Instruction FollowingLiveBench
77.11%
5 of 34
8%
LiveBench MathematicsLiveBench
85.82%
31 of 34
7%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it. Few tests yet: it has a job score of its own in only 1 of the 6 jobs, so the score may still move. It is listed, but gets no place.
GoogleClosed model
Gemini 3.1 Pro
Place#28of 35 models with enough tests
Silicon score70.168.3 to 71.9 with other tests
Price for developers$2 / $12in / out, per million tokens
Each job on its own · scored in 6 of 6
Chat & writing90.0
Coding62.3
Math & reasoning77.3
AI agents54.9
Web research89.0
Understanding images82.8
Its tests
Gemini 3.1 Pro took 25 of our tests, from 5 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
73.5%
6 of 35
9%
Arena SearchArena (LMArena)
1210 rating
13 of 17
9%
Arena AgentArena (LMArena)
-5.64 points
39 of 39
7%
τ³-bench Bankingτ³-bench (Sierra)
26.03%
12 of 12
7%
Arena WebDevArena (LMArena)
1446 rating
47 of 48
6%
MirrorCodeEpoch AI
8.89%
32 of 32
6%
Arena TextArena (LMArena)
1487 rating
15 of 41
5%
FACTS SearchFACTS (Google DeepMind and Kaggle)
83.7%
6 of 13
4%
LiveBench ReasoningLiveBench
84%
29 of 34
4%
EBR-benchEpoch AI
14.29%
18 of 21
4%
Arena VisionArena (LMArena)
1280 rating
17 of 30
4%
Arena DocumentArena (LMArena)
1443 rating
18 of 20
4%
Furniture AssemblyEpoch AI
26.67%
26 of 31
4%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
41.34%
8 of 13
4%
LiveBench LanguageLiveBench
85.38%
12 of 34
3%
LiveBench Instruction FollowingLiveBench
79.1%
3 of 34
3%
LiveBench CodingLiveBench
76.45%
30 of 34
3%
LiveBench Agentic CodingLiveBench
44.14%
34 of 34
3%
FrontierMath Tiers 1–3Epoch AI
59.65%
33 of 38
2%
Mystery Game PuzzlesEpoch AI
34%
23 of 36
2%
LiveBench MathematicsLiveBench
91.04%
24 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
65.67%
13 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
78.96%
3 of 14
1%
GPQA DiamondEpoch AI
94.44%
8 of 35
1%
Chess PuzzlesEpoch AI
51%
16 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
AnthropicClosed model
Claude Sonnet 5
Place#29of 35 models with enough tests
Silicon score69.767.6 to 71.8 with other tests
Price for developers$2 / $10in / out, per million tokens
Each job on its own · scored in 4 of 6
Chat & writing71.8
Coding79.0
Math & reasoning81.2
Web research75.6
Its tests
Claude Sonnet 5 took 18 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
33.7%
34 of 35
13%
Arena SearchArena (LMArena)
1188 rating
15 of 17
12%
Arena AgentArena (LMArena)
+1.14 points
25 of 39
10%
Arena WebDevArena (LMArena)
1539 rating
42 of 48
8%
MirrorCodeEpoch AI
63.33%
10 of 32
8%
Arena TextArena (LMArena)
1461 rating
34 of 41
7%
Arena VisionArena (LMArena)
1261 rating
24 of 30
6%
Arena DocumentArena (LMArena)
1460 rating
14 of 20
6%
LiveBench ReasoningLiveBench
88.69%
16 of 34
5%
LiveBench CodingLiveBench
80.68%
15 of 34
4%
LiveBench Agentic CodingLiveBench
59.39%
11 of 34
4%
LiveBench LanguageLiveBench
74.97%
30 of 34
3%
LiveBench Instruction FollowingLiveBench
63.86%
28 of 34
3%
FrontierMath Tiers 1–3Epoch AI
65.61%
31 of 38
3%
Mystery Game PuzzlesEpoch AI
35%
21 of 36
3%
LiveBench MathematicsLiveBench
92.94%
18 of 34
3%
GPQA DiamondEpoch AI
90.53%
29 of 35
1%
Chess PuzzlesEpoch AI
45%
23 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Thinking Machines LabOpen modelFew tests yet
Inkling 2 Small
Place–None yet: few tests yet
Silicon score69.265.6 to 72.8 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 0 of 6
It does not have enough tests in any single job for a job score of its own yet.
Its tests
Inkling 2 Small took 5 of our tests, from 2 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
Arena AgentArena (LMArena)
+2.51 points
19 of 39
30%
Arena WebDevArena (LMArena)
1645 rating
21 of 48
25%
Arena TextArena (LMArena)
1412 rating
41 of 41
21%
ArXivMathMathArena (ETH Zürich)
31.88%
7 of 7
12%
BrokenArXivMathArena (ETH Zürich)
63.19%
4 of 7
12%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it. Few tests yet: it has a job score of its own in only 0 of the 6 jobs, so the score may still move. It is listed, but gets no place.
AnthropicClosed model
Claude Haiku 5.5
Place#30of 35 models with enough tests
Silicon score68.866.7 to 70.9 with other tests
Price for developers$0.10 / $0.50in / out, per million tokens
Each job on its own · scored in 4 of 6
Chat & writing61.6
Coding81.6
Math & reasoning78.8
AI agents67.5
Its tests
Claude Haiku 5.5 took 17 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
23.8%
35 of 35
15%
Arena AgentArena (LMArena)
-0.37 points
31 of 39
11%
Arena WebDevArena (LMArena)
1593 rating
33 of 48
9%
MirrorCodeEpoch AI
78.89%
5 of 32
9%
Arena TextArena (LMArena)
1421 rating
40 of 41
8%
Furniture AssemblyEpoch AI
47.5%
14 of 31
7%
LiveBench ReasoningLiveBench
84.63%
28 of 34
6%
EBR-benchEpoch AI
43.81%
12 of 21
6%
LiveBench CodingLiveBench
77.86%
26 of 34
5%
LiveBench Agentic CodingLiveBench
57.17%
14 of 34
5%
LiveBench LanguageLiveBench
63.52%
34 of 34
4%
LiveBench Instruction FollowingLiveBench
66.47%
25 of 34
4%
FrontierMath Tiers 1–3Epoch AI
75.09%
20 of 38
3%
Mystery Game PuzzlesEpoch AI
30%
26 of 36
3%
LiveBench MathematicsLiveBench
92.75%
20 of 34
3%
GPQA DiamondEpoch AI
89.65%
34 of 35
2%
Chess PuzzlesEpoch AI
55%
14 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Z.ai (GLM)Open model
GLM-5.2
Place#31of 35 models with enough tests
Silicon score67.565.4 to 69.6 with other tests
Price for developers$1.40 / $4.40in / out, per million tokens
Each job on its own · scored in 4 of 6
Chat & writing73.9
Coding75.3
Math & reasoning68.5
AI agents62.7
Its tests
GLM-5.2 took 17 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
34.2%
33 of 35
14%
Arena AgentArena (LMArena)
-0.49 points
32 of 39
11%
τ³-bench Bankingτ³-bench (Sierra)
37.11%
9 of 12
11%
Arena WebDevArena (LMArena)
1608 rating
32 of 48
9%
MirrorCodeEpoch AI
36.67%
24 of 32
9%
Arena TextArena (LMArena)
1475 rating
26 of 41
8%
LiveBench ReasoningLiveBench
78.63%
33 of 34
6%
EBR-benchEpoch AI
9.52%
20 of 21
5%
LiveBench LanguageLiveBench
76.24%
28 of 34
4%
LiveBench Instruction FollowingLiveBench
62.29%
31 of 34
4%
LiveBench CodingLiveBench
79.65%
18 of 34
4%
LiveBench Agentic CodingLiveBench
51.77%
27 of 34
4%
FrontierMath Tiers 1–3Epoch AI
59.21%
34 of 38
3%
Mystery Game PuzzlesEpoch AI
19%
33 of 36
3%
LiveBench MathematicsLiveBench
89.78%
26 of 34
3%
GPQA DiamondEpoch AI
91.86%
21 of 35
1%
Chess PuzzlesEpoch AI
22%
34 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-5.6 Luna
Place#32of 35 models with enough tests
Silicon score66.764.7 to 68.7 with other tests
Price for developers$0.20 / $1.20in / out, per million tokens
Each job on its own · scored in 5 of 6
Chat & writing70.7
Coding70.2
Math & reasoning77.7
Web research76.8
Understanding images79.9
Its tests
GPT-5.6 Luna took 19 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
41%
30 of 35
13%
Arena WebDevArena (LMArena)
1519 rating
44 of 48
8%
MirrorCodeEpoch AI
21.11%
30 of 32
8%
Arena TextArena (LMArena)
1453 rating
37 of 41
7%
FACTS SearchFACTS (Google DeepMind and Kaggle)
79.35%
12 of 13
6%
LiveBench ReasoningLiveBench
85.64%
27 of 34
6%
Arena VisionArena (LMArena)
1255 rating
27 of 30
6%
Arena DocumentArena (LMArena)
1454 rating
15 of 20
6%
Furniture AssemblyEpoch AI
42.5%
17 of 31
6%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
35.81%
12 of 13
6%
LiveBench LanguageLiveBench
72.57%
33 of 34
4%
LiveBench Instruction FollowingLiveBench
60.12%
32 of 34
4%
LiveBench CodingLiveBench
82.92%
6 of 34
4%
LiveBench Agentic CodingLiveBench
48.43%
32 of 34
4%
FrontierMath Tiers 1–3Epoch AI
82.11%
17 of 38
3%
Mystery Game PuzzlesEpoch AI
21%
32 of 36
3%
LiveBench MathematicsLiveBench
87.2%
30 of 34
3%
GPQA DiamondEpoch AI
91.6%
23 of 35
1%
Chess PuzzlesEpoch AI
43%
24 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Z.ai (GLM)Open model
GLM-5.3 Flash
Place#33of 35 models with enough tests
Silicon score66.564.1 to 68.9 with other tests
Price for developers$0.15 / $0.50in / out, per million tokens
Each job on its own · scored in 3 of 6
Chat & writing73.0
Coding78.0
Math & reasoning62.2
Its tests
GLM-5.3 Flash took 14 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
Arena VisionArena (LMArena)
1272 rating
18 of 30
18%
Arena AgentArena (LMArena)
-3.7 points
37 of 39
14%
Arena WebDevArena (LMArena)
1609 rating
30 of 48
12%
Arena TextArena (LMArena)
1475 rating
24 of 41
10%
LiveBench ReasoningLiveBench
77.64%
34 of 34
8%
LiveBench CodingLiveBench
78.95%
19 of 34
6%
LiveBench Agentic CodingLiveBench
56.77%
16 of 34
6%
LiveBench LanguageLiveBench
77.31%
27 of 34
5%
LiveBench Instruction FollowingLiveBench
52.82%
34 of 34
5%
FrontierMath Tiers 1–3Epoch AI
55.79%
37 of 38
4%
Mystery Game PuzzlesEpoch AI
8%
35 of 36
4%
LiveBench MathematicsLiveBench
81.24%
34 of 34
4%
GPQA DiamondEpoch AI
90.15%
32 of 35
2%
Chess PuzzlesEpoch AI
21%
35 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
OpenAIClosed model
GPT-6 Luna
Place#34of 35 models with enough tests
Silicon score64.762.8 to 66.6 with other tests
Price for developers$0.10 / $0.50in / out, per million tokens
Each job on its own · scored in 5 of 6
Chat & writing71.3
Coding72.3
Math & reasoning72.9
Web research73.4
Understanding images73.1
Its tests
GPT-6 Luna took 22 of our tests, from 4 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
41.4%
29 of 35
11%
Arena SearchArena (LMArena)
1160 rating
17 of 17
10%
Arena VisionArena (LMArena)
1235 rating
30 of 30
10%
Arena AgentArena (LMArena)
+0.79 points
28 of 39
8%
Arena WebDevArena (LMArena)
1577 rating
37 of 48
7%
MirrorCodeEpoch AI
26.67%
28 of 32
7%
Arena TextArena (LMArena)
1441 rating
39 of 41
6%
FACTS SearchFACTS (Google DeepMind and Kaggle)
81.42%
8 of 13
5%
Furniture AssemblyEpoch AI
44.17%
15 of 31
5%
FACTS MultimodalFACTS (Google DeepMind and Kaggle)
34.25%
13 of 13
5%
LiveBench ReasoningLiveBench
81.77%
31 of 34
4%
LiveBench LanguageLiveBench
73.83%
32 of 34
3%
LiveBench Instruction FollowingLiveBench
55.93%
33 of 34
3%
LiveBench CodingLiveBench
78.95%
19 of 34
3%
LiveBench Agentic CodingLiveBench
51.21%
28 of 34
3%
FrontierMath Tiers 1–3Epoch AI
78.95%
19 of 38
2%
Mystery Game PuzzlesEpoch AI
7%
36 of 36
2%
LiveBench MathematicsLiveBench
89.12%
28 of 34
2%
FACTS GroundingFACTS (Google DeepMind and Kaggle)
81.37%
4 of 15
1%
FACTS ParametricFACTS (Google DeepMind and Kaggle)
56.08%
9 of 14
1%
GPQA DiamondEpoch AI
90.47%
31 of 35
1%
Chess PuzzlesEpoch AI
36%
31 of 37
1%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.
Moonshot AI (Kimi)Open model
Kimi K2.6
Place#35of 35 models with enough tests
Silicon score63.861.7 to 65.9 with other tests
Price for developersNo token priceno official price per token
Each job on its own · scored in 4 of 6
Chat & writing72.3
Coding68.7
Math & reasoning66.1
Understanding images79.0
Its tests
Kimi K2.6 took 17 of our tests, from 3 publishers. Its place is among the models we list that took the same test. The share says how much of the score each test decides.
Test
Result
Place
Share
SimpleQA VerifiedEpoch AI
34.9%
32 of 35
16%
Arena WebDevArena (LMArena)
1510 rating
46 of 48
10%
Arena TextArena (LMArena)
1461 rating
33 of 41
8%
LiveBench ReasoningLiveBench
79.38%
32 of 34
7%
Arena VisionArena (LMArena)
1265 rating
20 of 30
7%
Arena DocumentArena (LMArena)
1450 rating
17 of 20
7%
Furniture AssemblyEpoch AI
21.67%
29 of 31
7%
EBR-benchEpoch AI
2.38%
21 of 21
6%
LiveBench CodingLiveBench
78.57%
23 of 34
5%
LiveBench Agentic CodingLiveBench
46.92%
33 of 34
5%
LiveBench LanguageLiveBench
75.14%
29 of 34
4%
LiveBench Instruction FollowingLiveBench
64.36%
27 of 34
4%
FrontierMath Tiers 1–3Epoch AI
57.19%
35 of 38
3%
Mystery Game PuzzlesEpoch AI
18%
34 of 36
3%
LiveBench MathematicsLiveBench
84.28%
33 of 34
3%
GPQA DiamondEpoch AI
90.78%
26 of 35
2%
Chess PuzzlesEpoch AI
20%
37 of 37
2%
All its tests are put together in one score, and each of the 6 jobs counts the same, so the total is not a plain average of the job scores. The best model gets 100, and every 10 points less is one typical gap behind it.