CritPt Leaderboard

CritPt (Complex Research using Integrated Thinking - Physics Test (CritPt)): CritPt (Complex Research using Integrated Thinking - Physics Test; reads as 'critical point') tests LLMs on unpublished, research-level physics reasoning across ~12 subfields, with machine-verifiable numeric/symbolic answers designed to resist guessing. Frontier models score in the single digits (best base model GPT-5 high approx 5.7%, approx 10% with coding tools).

GPT-5.6 Sol leads CritPt at 32.3. The full leaderboard above lists every recorded measurement, not just the headline number.

76 models have recorded scores on CritPt, spanning a score spread of 32.3.

tensor.news grades CritPt A for integrity (score 94/100), ranking #23 of 61 benchmarks we assess across discrimination, saturation, contamination resistance, harness comparability, and freshness.

No — CritPt still has headroom and continues to discriminate between models rather than bunching them at the ceiling.

Its test set is held-out, which lowers — but does not eliminate — the risk that training data leaked into the questions.

Scores are reported under a consistent harness, so comparisons on CritPt are reasonably apples-to-apples.

Every CritPt measurement is source-backed and tagged reproduced, self-reported, or unverified; the underlying record cites arxiv.org/abs/2509.26574.