New to this? Read it in simple words
- Anthropic released a new AI model called Claude Sonnet 5.5.
- It costs half as much per token as Anthropic’s top model and scores almost as well in its tests.
- An independent tester agrees that it is nearly as good.
- But at the highest setting it used many tokens per task, which made it expensive.
- Token
- A small piece of text, often part of a word, that AI companies count and charge for.
- Effort setting
- How long a model may think; more effort usually means better answers and a higher cost.
- Benchmark
- A standard test used to compare AI models.
Anthropic released Claude Opus 5.5 on September 22. Six days later it added Sonnet 5.5, a faster and cheaper model in the same family.
What Anthropic claims
Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens, the same as Sonnet 5 and half the price of Opus 5.5.
Anthropic says it generates output more than 30% faster than Sonnet 5 and costs up to 30% less per task, because it needs far fewer tokens.
On Anthropic’s own tests, it scores 70.6% on Terminal-Bench 4.0, an agentic coding test, against 66.4% for Opus 5.5 and 10.3% for Sonnet 5. On GDPval-AA, a test of office work across 44 professions, it scores 1,844 against 1,846 for Opus 5.5.
Anthropic’s claims next to Artificial Analysis’s independent results.
What the independent test found
Artificial Analysis gave Sonnet 5.5 a score of 56 on its Intelligence Index at maximum effort, two points behind Opus 5.5.
In its launch-day test at that setting, it used about 193,000 output tokens per task, the most the firm had measured. As of October 8, Artificial Analysis lists $5.46 per task, about 7% more than Sonnet 5 at its highest setting.
Anthropic’s savings claims mostly refer to lower effort settings, where costs are much smaller: $0.48 per task at medium effort, by Artificial Analysis’s current count. Its launch-day test used a pre-release version with a since-fixed bug, and gave higher costs.
Safeguards and what comes next
Because its cyber skills are comparable to Opus 5’s, Anthropic says, Sonnet 5.5 is the first Sonnet to launch with cyber safeguards. Higher-risk cybersecurity requests fall back to Sonnet 5.
Artificial Analysis saw that fallback in about 0.1% of its tasks.
Sonnet 5.5 is available on Anthropic’s platform, AWS, Google Cloud and Microsoft Azure. Anthropic says a cheaper Haiku 5.5 will follow in the coming weeks.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2Research · September 28, 2026Claude Sonnet 5.5 reaches #2 on the Artificial Analysis Intelligence Index Artificial Analysis
- 3Research · September 28, 2026Anthropic’s Claude Sonnet 5.5 nearly matches Opus 5.5 on benchmarks while costing up to 30 percent less per task The Decoder
- 4Research · September 28, 2026Anthropic launches Claude Sonnet 5.5 with 30% cost reduction per-task due to faster speeds and fewer tool calls VentureBeat
- 5Research · Checked October 8, 2026Claude Sonnet 5.5 (max): Intelligence, Performance & Price Analysis Artificial Analysis
- 6Research · Checked October 8, 2026Claude Sonnet 5.5 (Medium): Intelligence, Performance & Price Analysis Artificial Analysis
- 7Research · Checked October 8, 2026Claude Sonnet 5 (Max): Intelligence, Performance & Price Analysis Artificial Analysis
The speed and cost-per-task claims are Anthropic’s. Some scores Anthropic cites were run for it by Artificial Analysis on a pre-release version. Artificial Analysis’s costs are those it listed on October 8 and can change when it tests again.
- We first gave Artificial Analysis’s launch-day costs, from its test of a pre-release version: $7.60 per task at maximum effort, about 50% more than Sonnet 5, and $0.59 at medium effort. It now lists $5.46, about 7% more than Sonnet 5, and $0.48.
Anthropic’s Sonnet 5.5 nears Opus at half the token price. Silicon AI News, September 29, 2026. https://siliconainews.com/stories/claude-sonnet-5-5-launch
Anthropic launched Sonnet 5.5 close to Opus at half the token price, though an independent test found higher costs at top effort. How we rate the mood
The mood shows how the news frames the event for the tech and AI world: as an opportunity (6 and above) or as a loss (below 5). It is not our verdict on who is right.