TechCrunch: OpenAI launches GPT-6.1 Sol, says it nearly matches GPT-6 Astra and costs less
At DevDay 2026, OpenAI released GPT-6.1 Sol, a mid-tier model positioned between the high-end GPT-6 Astra and the lightweight models in the GPT-4o family. OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra on benchmark performance while costing roughly one-fifth as much per token.
The accuracy figures are the most directly relevant detail for writing workflows. GPT-6.1 Sol has a factual error rate of 7.7%, down from 11.4% in previous mid-tier releases. For writers using AI to draft research-heavy content, summaries, or first-pass factual sections, that drop matters: fewer fabricated citations, fewer incorrect statistics to catch in review. The model also shows improvements in instruction-following, which affects tasks like holding a consistent style guide, applying specific formatting rules, or maintaining a constrained voice across a long document.
GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise, and Edu subscribers through ChatGPT, and through the API for developers. Notably, GPT-6.1 Astra — the version above Sol — was quietly scrapped before release for safety reasons. OpenAI cited elevated scores on deception-related evaluations as the reason, which is a separate and significant data point: the company is treating deception as a threshold condition rather than a tradeoff to manage, at least at this model tier.
For content teams, the practical implication is that a model with near-flagship accuracy is now available at a price point that makes large-scale document processing feasible. Batch jobs — generating first drafts across a content calendar, processing interview transcripts, running editorial summaries — that were cost-prohibitive with GPT-6 Astra become more viable with GPT-6.1 Sol. Whether the remaining 7.7% error rate is acceptable depends entirely on the verification workflow on the other end.