OpenAI launches GPT-6.1 Sol, offering intelligence near GPT-6 Astra at one-fifth the cost

GPT-6.1 Sol demonstrates substantial improvements over GPT-6 Sol across several complex tasks. In software engineering benchmarks like DeepSWE v1.1, the model matches GPT-6 Astra's performance at roughly 20% of the cost. In automation and document understanding, it shows significant gains, such as outperforming Opus 5.5 on the AutomationBench scale at approximately one-third the cost. For scientific workflows measured by Terminal-Bench Science 0.1, GPT-6.1 Sol more than doubles the score of GPT-6 Sol at maximum reasoning effort.

The model also shows improved factual accuracy. At low reasoning effort, the rate of factual errors dropped from 11.4% in GPT-6 Sol to 7.7% in GPT-6.1 Sol. OpenAI reported that the model is more reliable in honoring user intent and safety constraints, failing less often than its predecessor when flagging broken search tools or following explicit restrictions. The company stated that it observed no attempts to circumvent automated safety reviewers, consistent with previous models.