FeynmanBench - Momentum Routing: leaderboard
FeynmanBench (2026) gives a multimodal model a machine-generated Feynman diagram (Standard Model QED, electroweak or QCD, tree level to two loops) with minimal labelling conventions and asks it to recover the full physical content of the diagram. Checkpoint 4: an internal momentum assignment that conserves momentum at every vertex, judged by a three-model LLM panel; pass rate in percent, higher is better.
Metric: CP4 pass rate (%), a self-consistent internal momentum assignment conserving momentum at every vertex, judged by a three-model LLM panel, on FeynmanBench's Feynman-diagram tasks (a rendered diagram plus minimal conventions; Standard Model QED, electroweak and QCD diagrams from tree level to two loops), API models at temperature 0, mean of five runs; four models ran on subsets of 165 to 742 items, the others on about 2,000; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 17 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 23.9 | #102 |
| 2 | Claude Sonnet 4.6 | 12.6 | #179 |
| 3 | Qwen 3.6 Plus | 9.7 | #245 |
| 4 | Qwen 3.5 Plus | 9.4 | #250 |
| 5 | GLM-5V Turbo | 7 | #365 |
| 6 | Seed 2.0 Pro | 5.4 | #226 |
| 7 | Gemini 3 Flash | 5 | #180 |
| 8 | Claude Opus 4.7 | 4.7 | #86 |
| 9 | Claude Opus 4.6 | 4.5 | #114 |
| 10 | GPT-5.4 | 2.1 | #138 |
| 11 | GPT-5.1 | 1.3 | #248 |
| 12 | GPT-5.2 | 0.9 | #190 |
| 13 | Claude Opus 4.5 | 0.9 | #165 |
| 14 | Claude Sonnet 4.5 | 0.6 | #267 |
| 15 | Qwen 3.6 Max | 0.5 | #192 |
No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=feynmanbench-momentum-routing · How It Works · Data refreshed daily, snapshot 2026-11-22.