pass@1 dead even, but Fable 5.1 wins pass@k over GPT 6 Astra

zainhas · x · 2026-09-05

Follow-up data from the same Terminal Bench 4.0 runs: GPT 6 Astra and Fable 5.1 are dead even on pass@1, but GPT 6 Astra is less impressive at pass@k for k>1, suggesting less headroom from multi-attempt sampling.

Original post →

More from Models

Models channel →