CLAIM-2026-09-23-02
GPT-6 Sol scores 68.8% on DeepSWE 1.1 at maximum reasoning effort, within 1.1 points of Claude Fable 5 at 69.9%, at 80% lower cost, per OpenAI's own announcement.
Why it matters
It is the number the whole market will quote, and it is a max-effort number - which is not the setting most teams actually ship.
Sources
- openai2026gpt
- venturebeat2026openai