Reflection Beam Safety - ABC-Bench Screening Evasion - Best-of-3: leaderboard
Computational biological sequence-screening agent tasks,10 sequences,jury rubric score. Average and best-of-three are distinct; timeout counts remain in provenance. Source-specific score; no fusion with a different harness or version is implied. Computational sequence task rubric score; publisher excludes timeouts. Refusals do not change denominator for admitted metric. Metric: Best-of-3 (%). Higher is better. Exact printed figure values; https://cdn.sanity.io/images/sp40emik/production/24fb85302de02b9e6f6fe8a4ec7d3e44c147d1ae-2880x1998.png?w=3840&fit=max
Metric: Best-of-3 (%). Source: reflection.ai. Saturation forecast: Around April 2028. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K3 | 65.5 | #29 |
| 2 | Qwen 3.8 Max | 65.02 | #31 |
| 3 | GLM-5.3 | 63.27 | #40 |
| 4 | GLM-5.2 | 61.17 | #71 |
| 5 | Beam 1.0 | 51.57 | #97 |
| 6 | GPT-OSS-120B | 51.13 | #336 |
| 7 | Muse Glimmer 30B | 51.03 | #178 |
| 8 | Kimi K2.6 | 49.05 | #105 |
| 9 | DeepSeek V4 Pro | 48.02 | #101 |
| 10 | DeepSeek V4 Flash | 46 | #135 |
| 11 | Inkling | 43.8 | #123 |
| 12 | Nemotron 3 Ultra | 40.53 | #240 |
| 13 | DeepSeek V4.1 Flash | 38.3 | #35 |
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-abc-bench-screening-evasion-best-of-3 · How It Works · Data refreshed daily, snapshot 2026-10-17.