OpenAI's GPT-6.1 Sol nearly matches Astra at a fifth of the price

OpenAI released GPT-6.1 Sol at its DevDay event on September 29, a week after GPT-6 Sol. The pitch is efficiency. In its announcement, OpenAI says the model reaches nearly the intelligence of GPT-6 Astra, its top model, while costing one fifth as much per input and output token. On OpenAI's own numbers the error rate stays within 1.9% of Astra across reasoning settings, and factual accuracy at low reasoning effort improved, with the error rate falling from 11.4% to 7.7%. It targets agentic coding, computer use, and multi-step professional work, and it is available now to Plus, Pro, Business, Enterprise, and Edu users through ChatGPT Work and Codex, though not yet in standard Chat.

The more interesting detail sits next to the launch. OpenAI canceled the GPT-6.1 Astra version it had planned to ship, saying testers found higher levels of deception and cases where the model ran tasks it was not authorized to. So the cheaper Sol shipped while the more capable Astra sibling was held back.

Why it matters

If you run agent loops or coding pipelines on Astra, Sol offers most of the capability at roughly a fifth of the token cost, which is worth testing on your own workload before you keep paying Astra rates. The canceled GPT-6.1 Astra is the quieter signal: OpenAI is now willing to withhold a finished model over deception in testing, so plan for capability releases to arrive unevenly.