OpenAI announced GPT-6.1 Sol on September 29, 2026, as an upgrade to GPT-6 Sol for coding, computer use, professional work and multi-step business workflows. OpenAI lists API rates of $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens, and reports results close to GPT-6 Astra on several specific evaluations.
OpenAI announces GPT-6.1 Sol
OpenAI positions GPT-6.1 Sol for work involving code, computer use, complex documents and multi-step business workflows. Its API model identifier is gpt-6.1-sol.
The company’s performance claims are tied to named benchmarks and test conditions. They describe those evaluations, not a general result across every task.
Standard API rates and initial access
OpenAI’s listed standard API rates are $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Cached input refers to input tokens that qualify for the cached rate; output tokens are billed separately.
At the time of the announcement, OpenAI said eligible Plus, Pro, Business, Enterprise and Edu users could access GPT-6.1 Sol in ChatGPT Work and Codex. Developers could use it through the API. OpenAI said it was not yet available in the main ChatGPT chat at that time.
What OpenAI’s benchmark comparisons report
On DeepSWE v1.1, OpenAI reported that GPT-6.1 Sol matched GPT-6 Astra at roughly one-fifth the cost. The company also said Sol scored 6.4 percentage points above GPT-6 Sol’s best result, using lower reasoning effort and cost.
On AutomationBench at medium reasoning effort, OpenAI reported that Sol scored 2.2 percentage points above Opus 5.5 at roughly one-third the cost. At the same setting, it reported a 4.8-point advantage over GPT-6 Sol.
On the offline set in OSWorld 2.0, OpenAI said Sol came within 2.1 percentage points of Astra at maximum reasoning effort, at roughly one-seventh the cost per task. It also reported Sol scoring seven percentage points above GPT-6 Sol on that set.
For Terminal-Bench Science 0.1 at maximum reasoning effort, OpenAI reported these average costs per task:
| Model | Average cost per task |
| GPT-6.1 Sol | $5.47 |
| Opus 5.5 | $23.21 |
| GPT-6 Astra | $23.80 |
OpenAI also reported that GPT-6 Astra had the highest score in that evaluation, at 68.1%.
A creator-run Stonelabs test
A separate nine-task Stonelabs evaluation reported a weighted score of 96 for GPT-6.1 Sol and a total test-suite cost of $2.63. The run included a failed “penguin” task, so its overall score sits alongside at least one reported miss.
Factuality results and the planned Ultrafast version
At low reasoning effort on deliberately difficult factuality prompts, OpenAI reported that responses containing at least one factual error fell from 11.4% for GPT-6 Sol to 7.7% for GPT-6.1 Sol. OpenAI said the prompts were not representative of typical use.
In its September 29 announcement, OpenAI also said it planned to offer an Ultrafast version for Codex in the coming days, claiming it would generate tokens up to eight times faster than standard speed.
