What's this about?
OpenAI's own comparison table shows Anthropic's Fable 5.1 still ahead on the Artificial Analysis Intelligence Index (65.7 vs 61.2), Humanity's Last Exam with tools (65.0% vs 57.2%), and the Coding Agent Index.
What exactly are these indexes?
Third-party composite benchmarks that average many tests — harder to cherry-pick than any single number, which is why "who leads the index" carries weight even when individual benchmarks disagree.
Why is this worth noting?
Even in its own launch material, OpenAI isn't claiming a clean sweep — Astra's profile is jagged: dominant on computer use and cyber, behind on general-intelligence composites and coding indexes. "Which model is best" now genuinely depends on the task, and the honest table quietly concedes it.