Edition·2026-11-19 · Thu
Fed from·Epoch AI·HF Open LLM·OpenAlex·OpenAI evals
Every reported score is tagged by how it was checked, and every benchmark is graded on the signals that decide whether a number still means something.
Latest measurement·2026-11-17·13 of 63 benchmarks graded A
How trustworthy the evaluations are
Trust holds the top-left; the bottom-right is where scores max out and belief erodes. The best-graded benchmarks:
A | FrontierMath-Tiers-1-3-v2-Private | consistent harness |
A | FrontierMath-Tier-4-v2-Private | consistent harness |
A | Mystery Game Puzzles | consistent harness |
A | MirrorCode | consistent harness |
A | DTBench | consistent harness |
A | SimpleBench | consistent harness |
Top 10 models by rank
| # | Model | Lab | Leads | Avg gap | Evidence |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2 | Anthropic | 13 | 1.9 | third-party |
| 2 | GPT-6 Astra | OpenAI | 5 | 7.9 | third-party |
| 3 | Claude Fable 5 | Anthropic | 4 | 17.5 | reproduced |
| 4 | Claude Opus 5.5 | Anthropic | 4 | 9.2 | third-party |
| 5 | GPT-5 | OpenAI | 4 | 36.4 | reproduced |
| 6 | Claude Fable 5.1 | Anthropic | 4 | 13.3 | third-party |
| 7 | GPT-6.2 Astra | OpenAI | 6 | 4.6 | unverified |
| 8 | GPT-5.5 | OpenAI | 4 | 21.8 | reproduced |
| 9 | GPT-6.1 Sol | OpenAI | 3 | 11.4 | third-party |
| 10 | Claude Sonnet 5.5 | Anthropic | 3 | 14.2 | third-party |
Leads = benchmarks topped; avg gap = mean points behind SOTA elsewhere. Evidence tags how each score was checked — a score is f(model, harness, protocol), not f(model).
Recent measurements
Reported scores — protocols may differ. How we compare
Last refresh
scores that contradicted an earlier claim in the last refresh.
Answer pages
Analysis
Head-to-head
Follow the record
Every new score, contradiction, and analysis as it lands — plain Atom, open in any feed reader. No account, no inbox.
Section map
ranked by evidence
graded on integrity
head-to-head matchups
developers tracked
what each benchmark measures
source records
how we rank & grade