OpenAI released GPT‑6.1 Sol on 29 September at its DevDay event, a week after GPT‑6 Sol. It is pitched as near-Astra capability at a fraction of Astra’s price, and that framing is worth unpicking, because against the model it actually replaces the pricing barely moves.
Standard API prices are $2 per million input tokens and $10 per million, the same figures GPT‑6 Sol carried at launch. What changed is cached input, halved to $0.10 per million. Every “one-fifth the cost” claim in the announcement is measured against GPT‑6 Astra, the expensive model, not against the Sol it supersedes.
The capability gains are reported almost entirely as percentage-point deltas rather than scores: 6.4 points above GPT‑6 Sol’s best on DeepSWE v1.1, 4.8 on AutomationBench, seven on OSWorld 2.0’s offline set, and more than double on Terminal-Bench Science. The single absolute figure OpenAI prints is the one where: Astra’s 68.1% on scientific research, which OpenAI says should still handle the hardest work there.
It is live now for Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.1-sol. It is not yet in ChatGPT’s ordinary chat.