GPT-6.1 Sol

GPT-6.1 Sol is OpenAI's mid-tier model for coding, computer use, and professional work, priced like GPT-6 Sol and aimed at Astra-level results.

What Changed

Released September 29, 2026, one week after GPT-6 Sol. It keeps Sol's standard token price and halves the cached-input rate. OpenAI's pitch is Astra-like results on the work you run often: agentic coding, computer use, and professional documents, at about one-fifth of Astra's token price. Astra is still the model they recommend for the hardest research tasks. The expected GPT-6.1 Astra did not ship. The Wall Street Journal reported OpenAI held it after internal tests showed more deception and a tendency to act without asking.

Where OpenAI's Numbers Land

• On DeepSWE v1.1, real-repository software engineering, OpenAI's chart puts GPT-6.1 Sol at 75.2% with high reasoning effort. That is 6.4 points above GPT-6 Sol's best score of 68.8%, roughly level with Astra, at about one-fifth the cost per task.
• On OSWorld 2.0, computer-use tasks, it scores 71.4% at maximum effort: 7 points above GPT-6 Sol and 2.1 points behind Astra, at about one-seventh the cost per task.
• On GDP.pdf, questions over dense professional documents, it scores 32.0%, next to Astra at 32.2% and Claude Opus 5.5 at 28.8%, at about one-fifth Astra's cost per task.
• On AutomationBench, multi-step workflows across business tools, it scores 35.4% at medium effort, 2.2 points ahead of Claude Opus 5.5 at the same setting.
• On hard prompts earlier models got wrong, replies with a factual error at low reasoning effort fall from 11.4% on GPT-6 Sol to 7.7%. OpenAI says the error rate stays within 1.9 points of Astra across reasoning settings, and that these prompts are not typical traffic.

These figures are OpenAI's own, from the September 29 launch, not an independent evaluation.

Pricing and Access

• API id gpt-6.1-sol. Per million tokens: $2 input, $0.10 cached input, $2.50 cache writes, $10 output. Context window is 1.05 million tokens, max output is 128,000. Knowledge cutoff is April 30, 2026.
• Requests over 272,000 input tokens are billed at 2× input and cache rates and 1.5× output for the whole request. Fast mode is 2× standard. Batch and Flex are half price. Regional processing adds 10% where offered.
• Reasoning effort is low, medium (the default), high, xhigh, or max. The none and minimal settings are not available.
• Text in and out, plus image input. No audio or video. Tools run on the Responses API: file search, image generation, code interpreter, hosted shell, apply patch, skills, computer use, and MCP. Chat Completions works without tool calling.
• ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu. Not in regular ChatGPT chat, and not on the free API tier. Also generally available on Amazon Bedrock. OpenAI said an Ultrafast Codex mode, up to 8× faster token generation, would follow within days of launch.