OpenAI ships GPT-6.1 Sol on 29 September
OpenAI published Introducing GPT-6.1 Sol on 29 September 2026 (US time; evening of 29 September in Singapore). The post frames GPT-6.1 Sol as an upgrade to GPT-6 Sol. See OpenAI's 22 September Sol and Luna post and earlier Frontier notes on GPT-6 Sol and Luna. OpenAI says it nearly matches GPT-6 Astra on agentic coding, computer use and professional work at one-fifth of Astra's standard API input and output token prices.
API model id is gpt-6.1-sol. Standard list prices stay $2 per million input tokens and $10 per million output tokens. Cached input is $0.10 per million tokens. OpenAI says that cached rate is 95% below standard input pricing and 50% below GPT-6 Sol's cached input price.
Plus through Edu get Work and Codex access
GPT-6.1 Sol is available the same day to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI says it is not yet available in Chat. Developers can call it on the OpenAI API as gpt-6.1-sol.
OpenAI also says GPT-6.1 Sol Ultrafast will arrive in Codex in the coming days. It claims up to 8x faster token generation than standard Codex speed. No ship date or separate Ultrafast price appears in the post.
OpenAI's DeepSWE, GDP.pdf and OSWorld scores
All scores below are OpenAI's own reports from the 29 September post. They are self-reported. No independent reproduction is cited in the post.
On DeepSWE v1.1, OpenAI says GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth the cost. It also says the model beats GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning effort and cost.
On GDP.pdf, OpenAI says GPT-6.1 Sol scores higher than Opus 5.5 with fallbacks at less than half the cost per task across the tested reasoning settings. It also says the model approaches Astra at roughly one-fifth the cost per task.
On AutomationBench 1.0.6, agents use 47 tools across sales, marketing, operations, support, finance and HR. OpenAI says GPT-6.1 Sol scores 2.2 points above Opus 5.5 at medium effort at roughly a third of the cost, and 4.8 points above GPT-6 Sol at the same setting. The post notes that its Claude Fable 5.1 cost point omits fallbacks that ran on about 40% of tasks.
On OSWorld 2.0's offline set (partial reward, v2026.08.08), OpenAI says GPT-6.1 Sol beats GPT-6 Sol by seven points at max effort at less than half the cost. It also says the model comes within 2.1 points of Astra at roughly one-seventh the cost per task.
On Terminal-Bench Science 0.1 at maximum effort, OpenAI says GPT-6.1 Sol more than doubles GPT-6 Sol's score at less than half the cost per task. Average cost is $5.47 per task versus $23.21 for Opus 5.5 and $23.80 for Astra, OpenAI reports. The post still ranks Astra highest at 68.1% among the models it tested.
On a hard factuality set of chats where users flagged an earlier error, OpenAI says low-effort factual-error share falls from 11.4% on GPT-6 Sol to 7.7% on GPT-6.1 Sol. It says these prompts are not typical usage.
The post also points to an addendum to the GPT-6 Astra system card for GPT-6.1 Sol on OpenAI's Deployment Safety Hub for alignment details.
Chat is still closed and scores are vendor-run
GPT-6.1 Sol is not in Chat yet. Ultrafast has no calendar date in the post. Standard API prices match GPT-6 Sol's $2 input and $10 output list; the new number that matters for agents is the $0.10 cached input rate.
Every benchmark comparison is OpenAI's. Competitor scores in the post come from publicly available reports or OpenAI's own harness. The post warns that research-environment and API evals can differ from production ChatGPT because of system prompts, tools and effort settings.
Re-pin Work, Codex and API ids this week
If you already run gpt-6-sol in ChatGPT Work, Codex or the API, plan a side-by-side on your agent harness before you switch the default. Keep Astra on the hardest scientific or high-stakes paths until your own evals say otherwise.
Budget agents that reuse long context against the $0.10 cached input rate, and watch for Ultrafast in Codex if latency is the bottleneck. See earlier Sol pricing in OpenAI launches GPT-6 Sol and Luna. Treat sandbox and tool risk as in OpenAI's DNS sandbox escape pause.