GPT-6.1 Sol Costs $2 Input and $10 Output Per Million Tokens, and OpenAI Rates It Critical in Cybersecurity
A.I. / news
GPT-6.1 Sol Costs $2 Input and $10 Output Per Million Tokens, and OpenAI Rates It Critical in Cybersecurity
The September 29 model is priced at a fifth of GPT-6 Astra, while its own system card addendum puts it in the top Preparedness tier for cyber.
OpenAI released GPT-6.1 Sol on September 29 at $2 per million input tokens and $10 per million output tokens, one-fifth of GPT-6 Astra's price. The company's system card addendum classifies it as Critical in cybersecurity and High in biological and chemical capabilities under its Preparedness Framework.
OpenAI said the model is available to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. The weights are closed.
Price: where Sol sits against Astra
VentureBeat's Carl Franzen reported the per-million-token prices on September 29. Cached input for GPT-6.1 Sol is $0.10, half the $0.20 charged for GPT-6 Sol.
| Model | Input | Cached input | Output |
|---|---|---|---|
| GPT-6.1 Sol | $2 | $0.10 | $10 |
| GPT-6 Sol | $2 | $0.20 | $10 |
| GPT-6 Astra | $10 | $1 | $50 |
A new Ultrafast tier costs six times standard pricing and reaches up to 300 tokens per second, VentureBeat reported. OpenAI said it is up to 8 times faster in Codex and up to 6 times faster through the API. For GPT-6 Astra that works out to $60 input and $300 output per million tokens.
- GPT-6.1 Sol10 USD
- GPT-6 Sol10 USD
- GPT-6 Astra50 USD
- GPT-6 Astra Ultrafast300 USD
Source: VentureBeat, September 29, 2026, citing OpenAI pricing
Performance claims are OpenAI's own
OpenAI said that on DeepSWE v1.1, a software-engineering benchmark built on real codebases, GPT-6.1 Sol matches GPT-6 Astra at roughly one-fifth of the cost. It also said Sol beats GPT-6 Sol's best score by 6.4 percentage points at lower reasoning effort. Both figures are vendor-supplied.
VentureBeat reported that OpenAI's benchmarks put GPT-6.1 Sol above Anthropic's Claude Opus 5.5 on professional document analysis. No outside party is named as having reproduced that comparison.
What the system card says about cyber
The addendum lists ExploitBench at a 99.7% success rate at maximum reasoning effort. On the harder ExploitBench Internal Port task, the arbitrary-code-execution rate is 21.5%, against 31.5% for GPT-6 Astra. ExploitGym shows 35.1% on the intended vulnerability.
The card also gives biology results: 55.34% pass@1 on Multimodal Troubleshooting Virology, 40.74% on ProtocolQA Open-Ended and 47.96% on TroubleshootingBench. OpenAI said Sol scores higher than GPT-6 Sol in 5 of 8 production safety categories.
The pairing is the point. The model OpenAI prices at a fifth of Astra is the one its own card places in the Critical cyber tier.
Health and other evaluations
OpenAI reported 64.2 on HealthBench Professional, length-adjusted, matching GPT-6 Astra. On MentalHealthBench the overall score is 57.9% plus or minus 1.0, slightly under Astra's 58.7%.
Context at OpenAI this month
The release follows OpenAI's account of a Moonshot-linked cluster replaying encrypted reasoning. It also lands after Wikimedia's report of OpenAI agents editing its wikis.
OpenAI's own announcement page returned an access error when The Terminal tried to read it, so the prices here come from VentureBeat and the safety figures from the addendum.
Sources
More in A.I.
- 01Reflection Announces Beam, a 501B Open-Weight Model, but Has Not Named a LicenceThe weights are promised for later in October. Reflection's own post gives scores and training scale, and says nothing on licence terms or API pricing.
- 02Meta Disputes Inc. Columnist's Claim That Muse Read His Messages While Mac Researcher Calls It a BackdoorTwo separate Muse problems are being reported as one. A researcher's local-access flaw is documented; the claim that the agent read private messages is contested by Meta.
- 03Anthropic Staff Reported a Claude Chat to Police, and a Florida Woman Faces a Felony ChargeThe company's privacy policy allows disclosure to prevent serious harm. Its transparency report counted zero emergency requests from police, and does not count referrals it makes itself.
- 04Vals AI's 90 Claude Opus 5.5 Agents Name Two Magnetic Semiconductor Candidates, Neither Yet MeasuredOne candidate was first synthesised in 1999 and has a measured ordering temperature of 376 K. The other may not survive the furnace.