OpenAI has released GPT-6.1 Sol, an upgrade that it says approaches GPT-6 Astra on coding, computer use, and professional work at one-fifth of Astra’s standard input and output token prices. The API price is $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. Cached input is half the price of GPT-6 Sol’s cached input.
The new model is available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. Developers can call it through the API as gpt-6.1-sol. OpenAI says it is not yet available in Chat. An Ultrafast option for Codex, with up to eight times faster token generation, is planned for the coming days.
Where the benchmark gains appear
On DeepSWE v1.1, OpenAI says GPT-6.1 Sol matches Astra’s score on complex software engineering tasks at roughly one-fifth the cost per task. It beats GPT-6 Sol’s best score by 6.4 percentage points while using a lower reasoning effort and costing less.
For work with complex PDFs, OpenAI reports that Sol scores above Opus 5.5 with fallbacks on GDP.pdf at less than half the cost per task across the tested settings. It approaches Astra’s score at roughly one-fifth the cost. On AutomationBench, Sol scores 2.2 percentage points above Opus 5.5 at medium effort for about a third of the cost, and 4.8 points above GPT-6 Sol at the same setting.
On the offline OSWorld 2.0 computer-use test, Sol scores seven points above GPT-6 Sol at maximum effort for less than half the cost. It finishes 2.1 points behind Astra at roughly one-seventh of Astra’s cost per task. These figures use the test’s partial reward measure, rather than a simple count of fully completed tasks.
Astra keeps a clearer lead on the hardest science work. At maximum effort on Terminal-Bench Science 0.1, it scores 68.1%, the highest of the models OpenAI tested. Sol more than doubles GPT-6 Sol’s score, but OpenAI does not give its exact score in the accompanying text. The average cost per task is $5.47 for Sol, $23.21 for Opus 5.5, and $23.80 for Astra. OpenAI recommends Astra for the most difficult scientific research tasks.
OpenAI also reports a lower factual error rate for Sol. At low effort, 7.7% of its answers contained an error, compared with 11.4% for GPT-6 Sol. Across tested efforts, Sol stays within 1.9 percentage points of Astra. The evaluation uses conversations where users had already flagged an earlier model’s error, so the rates describe a deliberately difficult set of prompts.
