NewsModels & ProductsNew
OpenAI releases GPT-6.1 Sol, says it nearly matches GPT-6 Astra at about a fifth of the price
At DevDay on 29 September, OpenAI released GPT-6.1 Sol for coding, computer use, and professional work; its performance figures are OpenAI's own and have not been independently reproduced.
Evidence: 4 sources · 2 official records · 2 independent reports · 3 open questions
What happened
- On 29 September, at its DevDay event, OpenAI released GPT-6.1 Sol, an upgrade to GPT-6 Sol.
- API pricing is listed at $2 per million input tokens and $10 per million output tokens, with cached input at $0.10.
- At launch it was available in the API, ChatGPT Work, and Codex for paid tiers, but not yet in regular ChatGPT chat.
- OpenAI's system card addendum says it treats the model as Critical for cybersecurity and High for biological and chemical capability under its Preparedness Framework.
Why it matters
Coding and agent work are where AI companies are competing hardest on price, and OpenAI's own safety ratings for this model are the highest it has published for cybersecurity.
What is confirmed and what is claimed
Confirmed
- GPT-6.1 Sol was released on 29 September during DevDay (OpenAI announcement; also reported by TechCrunch). 123
- Listed API prices are $2 input and $10 output per million tokens (OpenAI announcement; also reported by Dataconomy). 124
- OpenAI's deployment safety addendum rates the model Critical for cybersecurity and High for biological and chemical capability (OpenAI system card addendum). 12
Claimed, not independently verified
- OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra on agentic coding, computer use, and professional work at about one-fifth of Astra's standard token prices. 12
- OpenAI says the share of responses with a factual error on difficult prompts, at low reasoning effort, fell from 11.4% for GPT-6 Sol to 7.7%. 12
- OpenAI says the model matches GPT-6 Astra on the DeepSWE v1.1 coding benchmark. 12
OpenAI used its annual developer conference to release GPT-6.1 Sol, a mid-tier model it presents as close to its top GPT-6 Astra model on coding and agent tasks, at a much lower price. The same event introduced developer tools for agents, including computer use in the Agents API and cloud-based Codex environments.
Context
The headline comparison, "nearly matches Astra", rests on OpenAI's own evaluations. Company-run benchmarks can be useful signals, but they are chosen and run by the company, and small changes in settings such as reasoning effort can move results. Our guide to reading benchmark claims explains what to look for.
The pricing is the firmer fact: it is published, and customers can check it. A cheaper model that handles long coding and computer-use tasks matters most for developers who run many agent steps, where token costs add up quickly.
The safety rating is also notable. Under OpenAI's Preparedness Framework, the company itself classifies GPT-6.1 Sol as Critical for cybersecurity capability. That is a statement about what OpenAI believes the model could do, and it is the reason the system card spends time on safeguards.
What to watch
- Independent results from outside evaluators and public leaderboards.
- Whether and when the model reaches regular ChatGPT chat.
- Any public explanation of the cancelled GPT-6.1 Astra release.
What we don’t know yet
- How the model performs on independent evaluations run by outside groups.
- What safeguards accompany a model OpenAI itself rates Critical for cybersecurity, beyond what the system card describes.
- Why a GPT-6.1 Astra release did not go ahead; TechCrunch, citing the Wall Street Journal, reported it was scrapped over safety concerns raised in internal testing, which OpenAI has not detailed publicly.
Evidence and sources
- Primary announcement Introducing GPT-6.1 Sol
- Document or data Addendum to GPT-6 Astra System Card: GPT-6.1 Sol
- Independent reporting OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less
- Independent reporting OpenAI Launches GPT-6.1 Sol At DevDay