GPT 6.2 Astra tops Kevin Bryan's HardGeoBench as best OpenAI model yet, with caveats

Afinetheorem · x · 2026-11-20

Kevin Bryan updates his HardGeoBench image+logic localization benchmark with harness-and-style commentary: GPT 6.2 Astra is the best overall model and by far the strongest OpenAI model on the benchmark, despite being nearly unusable in other contexts. The model codenamed Neon is a great writer with swarm capabilities but burns 10x the tokens at 1/10th the speed of comparable models.

Original post →

More from Models

Models channel →