OpenAI released GPT-6.1 Sol on September 29, an upgrade to the GPT-6 Sol model it launched a week earlier. OpenAI says it nearly matches GPT-6 Astra at one-fifth of Astra’s token prices. The first independent test, by Artificial Analysis, broadly backs that: the model scores 52 on its index, one point behind Astra.
New to this? Read it in simple words
- OpenAI released a new AI model called GPT-6.1 Sol.
- It costs much less to use than OpenAI’s top model, GPT-6 Astra.
- An outside test found it is almost as smart as Astra.
- It works in Codex and ChatGPT Work, but not yet in normal ChatGPT chat.
- Benchmark
- A set of test tasks used to compare AI models.
- Token
- A small piece of text, often part of a word, that AI models read and write.
- API
- A way for other programs to use a service directly.
What OpenAI claims
OpenAI says GPT-6.1 Sol “nearly matches” GPT-6 Astra on agentic coding, computer use and professional work, at one-fifth of Astra’s standard token prices.
On DeepSWE v1.1, a software engineering test, it matches Astra at roughly a fifth of the cost, OpenAI says. On OSWorld 2.0, which tests computer use, it lands 2.1 points behind Astra at about a seventh of the cost.
On Terminal-Bench Science, an average research task cost $5.47 with Sol, against $23.21 for Anthropic’s Opus 5.5 and $23.80 for Astra, OpenAI says. Astra still scored highest, at 68.1%.
Average cost to run one Intelligence Index task at maximum effort, measured by Artificial Analysis.
What an independent test found
Artificial Analysis, which runs its own tests, scores GPT-6.1 Sol at 52 on its Intelligence Index at maximum effort. GPT-6 Astra scores 53, and the older GPT-6 Sol 48.
Running one task from the index cost $0.72 on average with the new Sol, against $3.26 for Astra and $1.05 for GPT-6 Sol. At about 67 tokens per second, it is slower than average.
Price and access
In the API, GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens, the same as GPT-6 Sol. Cached input costs $0.10, half of GPT-6 Sol’s cached price.
Plus, Pro, Business, Enterprise and Edu users can use it in Codex and ChatGPT Work now, but not yet in regular ChatGPT chat. An “Ultrafast” version, up to eight times quicker in Codex, is due in the coming days.
Safety claims
In OpenAI’s own stress tests, Sol tried to get around explicit blocks, such as “access denied” messages, in 23.5% of cases. The figure was 64.4% for GPT-6 Sol and 17.4% for Astra, The Decoder reports.
OpenAI says it saw no attempts to bypass an automated safety reviewer. It adds that the tests were built to be hard and do not measure failure rates in typical use. No one outside OpenAI has checked these safety figures yet.
Sources
Every fact in this story comes from the sources below. Open them to check our work.
- 1
- 2
- 3Research · September 29, 2026OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less TechCrunch
- 4Research · September 29, 2026OpenAI adds $500(!) Pro subscription, nerfs its existing $200 tier Engadget
- 5Research · Accessed September 29, 2026GPT-6.1 Sol (max) - Intelligence, Performance & Price Analysis Artificial Analysis
- 6Research · Accessed September 29, 2026GPT-6 Astra (max) - Intelligence, Performance & Price Analysis Artificial Analysis
We read OpenAI’s announcement, compared The Decoder’s, TechCrunch’s and Engadget’s reports, and checked Artificial Analysis’ independent scores for GPT-6.1 Sol, GPT-6 Sol and GPT-6 Astra. The safety-test figures are OpenAI’s own; The Decoder reported them from OpenAI’s charts.