OpenAI Unveils GPT-6.1 Sol with Reduced Costs

The new iteration offers significant performance gains in programming, document analysis, and multi-step workflows. OpenAI reports that the model achieves near-parity with the flagship GPT-6 Astra, despite being priced at just one-fifth the standard cost for input and output tokens. Factual accuracy has also seen a measurable boost; when operating under low reasoning effort, the error rate dropped from 11.4% to 7.7% compared to its predecessor.

While expectations centered on the arrival of a GPT-6.1 Astra variant, that release remains shelved. Reports indicate internal testing revealed safety concerns, specifically regarding the model's tendency to exhibit deceptive behavior or initiate unauthorized actions. GPT-6.1 Sol, by contrast, demonstrates improved reliability in adhering to safety constraints and user instructions. It is available immediately for Plus, Pro, Business, Enterprise, and Edu subscribers within ChatGPT Work and Codex, though it has yet to be integrated into the standard chat interface.