OpenAI releases GPT-6.1 Sol
Near-Astra performance at one-fifth of Astra’s price
Confirmed
Importance: major (4 of 5)The takeaway
On Sept 29, 2026, at DevDay and one week after GPT-6 Sol, OpenAI released GPT-6.1 Sol (API id gpt-6.1-sol).
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- Major (4 of 5)
- Last verified
- 10 October 2026
Your AI and this story
- GPT-6 Astra152 days after its cutoff
- Claude Opus 5.591 days after its cutoff
- Gemini 4 Hydrogen182 days after its cutoff
- Grok 4.7121 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 91 days before it.
Key facts
- API id gpt-6.1-sol (single snapshot); Responses API (tool calling), Chat Completions (no tool calling), Batch; text+image in, text out
- 1,050,000-token context window (922K max input), 128K max output, knowledge cutoff Apr 30, 2026; reasoning.effort low/medium (default)/high/xhigh/max (‘none’ and ‘minimal’ not supported)
- Price per 1M tokens: $2 input, $0.10 cached input (95% off; half of GPT-6 Sol’s $0.20), $2.50 cache writes, $10 output; >272K-token prompts cost 2x input and 1.5x output; Fast mode 2x, Batch/Flex 50% off
- Ultrafast version (up to 8x faster in Codex) promised ‘in the coming days’
- DeepSWE v1.1: matches GPT-6 Astra at ~1/5 the cost and beats GPT-6 Sol’s best score by 6.4 points (The New Stack reads it as ~75%)
- GDP.pdf (Surge AI): higher than Claude Opus 5.5 (with fallbacks) at under half the cost per task (The New Stack: ~32% vs ~29%)
- AutomationBench 1.0.6 (Zapier): +2.2 points over Opus 5.5 at medium effort at ~1/3 the cost; +4.8 over GPT-6 Sol
- OSWorld 2.0 offline: +7 points over GPT-6 Sol at max effort, within 2.1 points of Astra at ~1/7 the cost per task
Show 6 more
- Terminal-Bench Science 0.1: more than double GPT-6 Sol’s score; $5.47 per task at max effort vs $23.21 (Opus 5.5) and $23.80 (Astra); Astra still leads at 68.1%
- Factuality (hard, user-flagged prompts): responses with a factual error fell from 11.4% to 7.7% at low effort; within 1.9 points of Astra across settings
- Alignment: failed to disclose a broken search tool in 2.1% of test cases (GPT-6 Sol 4.9%, Astra 1.5%, GPT-6 Luna 28.7%); no attempts to bypass the automated safety reviewer observed; system card addendum published
- ARC-AGI (Verified, ARC Prize, Sept 30): ARC-AGI-3 52.7% ($7.6K, standard harness) and 96.4% ($4.4K, provider-adapter harness), close to GPT-6 Astra’s 99.9% at 77% lower cost; ARC-AGI-2 94.2% ($0.25/task); ARC-AGI-1 98.5% ($0.06/task)
- Availability: all Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex; GitHub Copilot Pro+, Max, Business and Enterprise (GA Sept 29); OpenRouter openai/gpt-6.1-sol (and -pro)
- Customer story (OpenAI, Oct 9): ‘Asana cuts model costs 76x in browser tests with GPT-6.1 Sol’. The page returned 403 to our fetcher, so only the headline figure is recorded; what the 76x is measured against is unverified
What happened
OpenAI shipped GPT-6.1 Sol at DevDay 2026, exactly one week after GPT-6 Sol (Sept 22). It calls the model “a major upgrade to GPT-6 Sol with exceptionally strong performance on agentic coding, along with computer use, and professional work”. The list price is the same as GPT-6 Sol ($2/$10 per 1M tokens), but cached input is halved to $0.10. That is one-fifth of GPT-6 Astra’s $10/$50. OpenAI’s benchmark charts compare it with Astra, GPT-6 Sol and Anthropic’s Claude Opus 5.5 (and in some footnotes Claude Fable 5.1). On most of them it lands close to Astra at a fraction of the cost per task. Astra still leads on the hardest scientific work (Terminal-Bench Science 68.1%).
The New Stack compared it with Anthropic’s Claude Sonnet 5.5, released a day earlier at the same $2/$10 list price. It found mixed results: DeepSWE about 75% vs 71% in Sol’s favour, but AutomationBench about 36% vs 44.7% for Sonnet, where Sol was much cheaper per task.
The model is available in ChatGPT Work and Codex, not yet in the regular Chat experience. GitHub made it generally available in Copilot the same morning.
Why it matters
Near-flagship capability is now available at mid-tier prices, a week after the previous mid-tier model. This continues the fast price compression of September 2026 (GPT-6 Sol/Luna at half price, Opus 5.5, Sonnet 5.5). It also shows that OpenAI’s next release after cancelling GPT-6.1 Astra was a cheaper model that OpenAI describes as better aligned, rather than a new flagship.
Sources
15 sources from 10 sites. Numbers match the chips in the text.
15 sources: 11 primary, 3 press, 1 reaction
Primary
- OpenAI: Introducing GPT-6.1 Solopenai.com, official
- GPT-6.1 Sol system card addendum (PDF)cdn.openai.com, paper
- OpenAI API docs: GPT-6.1 Sol model pagedevelopers.openai.com, docs
- OpenAI API pricingdevelopers.openai.com, docs
- GitHub Changelog: GPT-6.1 Sol in GitHub Copilotgithub.blog, official
- OpenAI on X: ‘GPT-6.1 Sol: near-Astra intelligence for a fifth of the price.’x.com, official
- OpenAI on X: GPT-6.1 Sol alignment evaluationsx.com, official
- OpenRouter: openai/gpt-6.1-solopenrouter.ai, docs
- ARC Prize on X: GPT-6.1 Sol on ARC-AGI (Verified)x.com, official
- ARC Prize: full GPT-6.1 Sol resultsarcprize.org, official
- OpenAI: Asana cuts model costs 76x in browser tests with GPT-6.1 Sol (Oct 9)openai.com, official
Press
- TechCrunch: OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astratechcrunch.com, press
- The New Stack: GPT-6.1 Sol undercuts its own Astra flagshipthenewstack.io, press
- The Decoder: GPT-6.1 Sol comes close to Astra at a fifth of the pricethe-decoder.com, press
Reactions
- Thibault Sottiaux on X: ‘an absolute workhorse’x.com, discussion
Changes
- Added OpenAI’s Asana customer story (headline only; page blocked)
- Added ARC Prize verified ARC-AGI-1/2/3 scores (Sept 30)
- Filed
