OpenAI has released the GPT-5.6 family, consisting of Luna, Terra, and Sol models, featuring a million-token context window and optimized for agentic workflows. The models demonstrate significant cost-efficiency improvements over competitors like Claude Fable 5, particularly in long-running professional tasks, although they lagged behind on specific coding benchmarks like SWE-Bench Pro.
Background
This release marks a significant update in OpenAI's flagship model lineup, emphasizing efficiency and agentic capabilities over raw benchmark scores. The pricing strategy highlights a shift towards valuing reasoning token efficiency rather than just input/output volume.
- Source
- Simon Willison
- Published
- Jul 10, 2026 at 03:46 AM
- Score
- 8.0 / 10