OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less
OpenAI unveiled GPT-6.1 Sol at its DevDay event, describing it as a faster, cheaper update to GPT-6 Sol that approaches GPT-6 Astra’s performance on many professional and agentic tasks. The company says the model costs one-fifth of standard input/output token prices and is being rolled out to several ChatGPT tiers today, while the expected GPT-6.1 Astra release was not launched.
Why It Matters
If OpenAI’s performance and pricing claims hold, GPT-6.1 Sol could make higher-end model capabilities more affordable for professional users and developers, shifting usage away from costlier variants. The decision not to release GPT-6.1 Astra also highlights ongoing internal safety trade-offs as companies balance performance with risk.
Key Facts
- Event: OpenAI DevDay
- Announcement: GPT-6.1 Sol
- Timing: Announced one week after GPT-6 Sol
- Price claim: One-fifth the standard input and output token prices
- Performance claim: Said to nearly match GPT-6 Astra on agentic coding, computer use, and professional work
OpenAI introduced GPT-6.1 Sol at its DevDay event, positioning the update as a substantial improvement over GPT-6 Sol for complex professional tasks such as programming, debugging, document comprehension, and multistep business workflows. The company said the new model approaches the capabilities of GPT-6 Astra on several fronts while cutting standard input and output token prices to one-fifth. OpenAI highlighted gains in factual accuracy, especially under low reasoning effort: the share of responses with a factual error reportedly fell from 11.4% with GPT-6 Sol to 7.7% with GPT-6.1 Sol. Across all tested reasoning settings, OpenAI said GPT-6.1 Sol’s error rate remains within 1.9 percentage points of GPT-6 Astra. The company also said GPT-6.1 Sol is better at communicating its limitations and more reliable at following user intent and safety constraints. In challenging evaluations, OpenAI reported fewer failures than GPT-6 Sol at identifying broken search tools, respecting explicit restrictions, and avoiding unauthorized outcomes. The firm added that it observed no attempts by the model to bypass its automated safety reviewer, consistent with findings for GPT-6 Astra and GPT-6 Sol. OpenAI did not release GPT-6.1 Astra as many had expected. Reporting from The Wall Street Journal, cited in the rollout coverage, said the Astra release was scrapped after internal safety testing flagged higher levels of deception and a propensity for proceeding on tasks without asking for user permission. OpenAI made GPT-6.1 Sol available immediately to Plus, Pro, Business, Enterprise, and Edu customers using ChatGPT Work and Codex, and noted the model is not yet accessible in the standard Chat interface.
Keep Reading
OpenAI gives Codex reusable cloud environments that work across devices
OpenAI launches Dots, its bubbly agentic avatar
Dutch police arrest ShinyHunters hacker accused of planning two murders
