OpenAI Launches GPT-6.1 Sol for Coding and Computer-Use Workloads

OpenAI introduced GPT-6.1 Sol on September 29, 2026 as an upgrade to GPT-6 Sol aimed at coding, computer use and other agentic professional workloads. It is available now to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, while developers can access it through the API as gpt-6.1-sol.

The rollout does not include ordinary ChatGPT conversations yet. OpenAI explicitly says GPT-6.1 Sol is not yet available in Chat, so availability in Work and Codex should not be interpreted as a replacement of the model used for every ChatGPT conversation.

OpenAI kept GPT-6.1 Sol's standard API prices at $2 per million input tokens and $10 per million output tokens while cutting cached input to $0.10 per million tokens. The company describes that cached-input rate as 95% below standard input pricing and 50% below GPT-6 Sol's cached-input price.

For long-running coding and agent workflows, cached-input pricing can matter because repeated instructions, tool definitions and project context may be reused across requests. The practical saving depends on an application's actual cache-hit behavior; the headline cached-token rate does not mean every input token receives the discount.

On OpenAI's reported DeepSWE v1.1 evaluation, GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth of the cost and exceeds GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort. On the OSWorld 2.0 offline set, OpenAI reports a seven-percentage-point gain over GPT-6 Sol at maximum reasoning effort, with GPT-6.1 Sol coming within 2.1 points of Astra at roughly one-seventh of the cost per task.

Those numbers are vendor-reported evaluation results, not independent Core Tech Tips benchmarks. OpenAI also notes that its GPT evaluations were performed in its research environment or API and may differ from production ChatGPT because system prompts, tools and reasoning settings can differ.

OpenAI says GPT-6.1 Sol reduces the share of responses containing at least one factual error from 11.4% with GPT-6 Sol to 7.7% at low reasoning effort, an approximately 32% relative reduction. Across the tested reasoning settings, OpenAI says the model remains within 1.9 percentage points of GPT-6 Astra.

That result should not be read as a general-purpose hallucination rate. OpenAI says the evaluation uses de-identified conversations where users had already flagged an earlier model error and that these deliberately difficult prompts are not representative of typical usage.

OpenAI says a GPT-6.1 Sol Ultrafast tier will arrive in the coming days with up to eight times faster token generation than standard speed in Codex. The announcement does not provide a universal tokens-per-second figure or claim that every workload receives an eightfold end-to-end speedup.

That distinction matters for coding agents because total task time also includes tool calls, compilation, tests, network operations and reasoning between generated tokens. Ultrafast therefore represents a model-serving speed option rather than a guaranteed eightfold reduction in complete workflow duration.