CyberBench v1.1 Patch
Agents
The 56-task patch track of CyberBench v1.1. A patch must compile, remove the original sanitizer failure and match maintainer-fixed reference behavior on hidden holdouts. This revised roster and clean-room grading protocol are distinct from the earlier 59-scored-item track.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Haiku 5.5 | 89.3 |
| GPT-5.6 Sol | 89.3 |
| GPT-6 Sol | 89.3 |
| DeepSeek-V4-Flash-0731 | 87.5 |
| Gemini 3.7 Flash | 87.5 |
| Gemini 3.8 Flash | 87.5 |
| GLM-5.3 | 87.5 |
| GPT-5.6 Terra | 87.5 |
| GPT-6 Luna | 87.5 |
| DeepSeek V4.1 Flash | 85.7 |
| DeepSeek V4 Pro 0813 | 85.7 |
| Ember-1 | 85.7 |
| Gemini 3.6 Flash | 85.7 |
| Gemini 4 Argon | 85.7 |
| GLM-5.3 Flash | 85.7 |
| Hy4 Preview | 85.7 |
| MiMo-V2.6-Flash | 85.7 |
| MiMo-V2.6-Pro | 85.7 |
| Muse Spark 1.2 | 85.7 |
| Muse Wonder | 85.7 |
| GPT-5.6 Luna | 83.9 |
| Grok 4.7 | 83.9 |
| Muse Wonder Full | 83.9 |
| Qwen3.8 27B | 83.9 |
| Sonnet 5 | 82.1 |
| GPT-6 Astra | 82.1 |
| Kimi K3 | 82.1 |
| Muse Spark 1.3 | 82.1 |
| Grok 4.5 | 80.4 |
| Grok 4.6 | 80.4 |
| GPT-6.1 Sol | 78.6 |
| Inkling | 78.6 |
| MiniMax M3 | 75.0 |
| Gemini 3.5 Flash-Lite | 73.2 |
| Inkling-Small | 73.2 |
| Mercury 2.5 | 64.3 |
| Nemotron 3.5 Lightning 30B A3B BF16 | 33.9 |