OpenAI announced GPT-6.1 Sol on September 29, 2026, as an upgrade to GPT-6 Sol for agentic coding, computer use and complex professional workflows. The company listed API access and access for eligible ChatGPT Work and Codex users, and published three API rates: $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens.

What OpenAI announced about GPT-6.1 Sol

OpenAI positioned GPT-6.1 Sol for work involving coding agents, computer use, document comprehension and multi-step business workflows. At launch, its access statement covered the API and eligible Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI said the model was not yet available in standard Chat.

That distinction matters if you use ChatGPT’s regular chat interface: the launch announcement named separate work and coding products, rather than standard Chat, as the places eligible subscribers could access Sol.

What the reported evaluations show

The footage shows code editors, terminal output and rendered web or 3D scenes from a custom nine-test run; its reported weighted score was 96.
The coding leaderboard displays model scores, execution times and costs alongside the reported medium-level result.

OpenAI reported results across several specific tests, with different models, tasks and reasoning settings as comparators. On DeepSWE v1.1, a software-engineering evaluation, OpenAI said GPT-6.1 Sol matched GPT-6 Astra’s score at roughly one-fifth the cost. The company also reported a 6.4-percentage-point gain over GPT-6 Sol’s best score, with lower reasoning effort and cost.

On AutomationBench at medium reasoning effort, OpenAI reported that Sol scored 2.2 percentage points above Opus 5.5 and 4.8 points above GPT-6 Sol at the same setting. It put Sol’s cost at about one-third of Opus 5.5’s for that evaluation. On the OSWorld 2.0 offline set at maximum reasoning effort, OpenAI reported a seven-point gain over GPT-6 Sol and a result 2.1 points below GPT-6 Astra; the task cost was roughly one-seventh of Astra’s.

OpenAI also reported that, in a low-effort factuality evaluation using deliberately difficult prompts, responses containing at least one factual error fell from 11.4% for GPT-6 Sol to 7.7% for GPT-6.1 Sol. The company said those prompts were not representative of typical use. These results describe particular evaluations and settings, not a single score for every kind of work.

Separate creator-run tests used their own coding tasks and scoring. In a nine-test Stonelabs suite, GPT-6.1 Sol received a weighted score of 96, tied with Sonnet 5.5 and ranked fourth; the reported total cost for the suite was $2.63. A different coding leaderboard reported 20/20 on medium-level tasks. Its higher-effort testing was incomplete, with prompts taking 6 to 9 minutes.

GPT-6.1 Sol API rates

OpenAI published these rates per million tokens. Cached input has its own lower rate, separate from standard input and output.

API chargePrice per million tokens
Standard input$2
Cached input$0.10
Output$10