FULL STORY
CritPt v2 Launch and the Score Revision Controversy
CritPt v2 launched with GPT-6 Astra topping the physics benchmark at 83%. Analysts then clarified that higher scores stem mainly from question revisions, making v1 and v2 results incomparable.
2026-10-22 ~ 2026-10-23 · 2 episodes · 6 posts
Episode 1 · CritPt v2 Physics Benchmark Released; GPT-6 Astra Tops at 83% (2026-10-22, 4 posts)
Artificial Analysis and the CritPt team released CritPt v2, a research-level physics benchmark of 70 revised expert-crafted problems. GPT-6 Astra topped the leaderboard at 83%, 54 points ahead of GPT-5, as frontier models approach research-level physics limits.
- Physics benchmark CritPt v2 launches: GPT-6 Astra leads at 83.1% accuracy — OfirPress · 2026-10-22
- CritPt v2 launches: GPT-6 Astra tops at 83%, 54 points above GPT-5 — ArtificialAnlys · 2026-10-23
- CritPt v2: GPT-6 Astra hits 83% on research-level physics benchmark as frontier closes in — ArtificialAnlys · 2026-10-23
- CritPt v2 launches: GPT-6 Astra tops research-level physics at 83%, 54 points above GPT-5 — ArtificialAnlys · 2026-10-23
Episode 2 · CritPt v2 Score Gains Driven by Prompt Revisions, Not Tooling (2026-10-23, 2 posts)
Artificial Analysis clarified that CritPt v2 scores are not directly comparable to v1, as the gains mainly come from prompt revisions rather than tooling—GPT-6 Astra (max) jumped from 32% to 83%.
- CritPt v2 gains mostly from revised challenges, not tool access — ArtificialAnlys · 2026-10-23
- CritPt v2 score jump mostly from revised challenges, not tool access, says Artificial Analysis — ArtificialAnlys · 2026-10-23