Reflection Beam Safety - WMDP-Chem - Log-likelihood: leaderboard
408 chemistry multiple-choice questions,zero-shot conditional likelihood in lm-eval,percentage correct. This is a knowledge score,not refusal behavior or a generated-answer test. Source-specific score; no fusion with a different harness or version is implied. All408 multiple-choice chemistry questions; zero-shot lm-eval conditional likelihood answer selection, not generated answers; no reasoning-effort label for this protocol. Metric: Accuracy (%). Higher is better. Exact printed figure values; https://cdn.sanity.io/images/sp40emik/production/5437f0a5b6c0120610c9ca66d0d59824f07cbaea-2280x1890.png?w=3840&fit=max
Metric: Accuracy (%). Source: reflection.ai. Saturation forecast: Around February 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 | 78.2 | #40 |
| 2 | GLM-5.2 | 77.5 | #71 |
| 3 | Kimi K3 | 77 | #29 |
| 4 | DeepSeek V4 Pro | 75.7 | #101 |
| 5 | Kimi K2.6 | 75.5 | #105 |
| 6 | Qwen 3.8 Max | 73.5 | #31 |
| 7 | DeepSeek V4 Flash | 73 | #135 |
| 8 | Nemotron 3 Ultra | 70.1 | #240 |
| 9 | Inkling | 65.6 | #123 |
| 10 | Muse Glimmer 30B | 63.7 | #178 |
| 11 | DeepSeek V4.1 Flash | 63.5 | #35 |
| 12 | Beam 1.0 | 62.3 | #97 |
| 13 | GPT-OSS-120B | 58.6 | #336 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=reflection-beam-safety-wmdp-chem-log-likelihood · How It Works · Data refreshed daily, snapshot 2026-10-17.