Reflection Beam Safety - Irregular CTF - Group - Atomic: leaderboard
Private50-task capture-the-flag suite in Mirror harness,autonomous shell use up to1000commands,exact planted flag grading. Difficulty,group andtype rows retain explicit disjoint task counts; score is solved percentage. Source-specific score; no fusion with a different harness or version is implied. Capability subset count/explicit catalog denominator; no subjective graph-distance digitization. Metric: Accuracy (%). Higher is better. Exact printed figure values; https://cdn.sanity.io/images/sp40emik/production/53ba143b4bbd4acb78e07ee7f81e6572edc52666-3000x2928.png?w=3840&fit=max
Metric: Accuracy (%). Source: reflection.ai. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 | 77.5 | #40 |
| 2 | DeepSeek V4 Pro | 75 | #101 |
| 3 | DeepSeek V4.1 Flash | 75 | #35 |
| 4 | Qwen 3.8 Max | 72.5 | #31 |
| 5 | GLM-5.2 | 70 | #71 |
| 6 | Kimi K3 | 70 | #29 |
| 7 | Beam 1.0 | 57.5 | #97 |
| 8 | Kimi K2.6 | 50 | #105 |
| 9 | Nemotron 3 Ultra | 45 | #240 |
| 10 | Inkling | 40 | #123 |
| 11 | Muse Glimmer 30B | 40 | #178 |
| 12 | GPT-OSS-120B | 15 | #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-irregular-ctf-group-atomic · How It Works · Data refreshed daily, snapshot 2026-10-17.