RuVerBench - Agentic Coding: leaderboard
LLM rubric verification of long research reports or coding trajectories, paper v2 Table 2. Overall is the macro-average category balanced accuracy. These are verifier scores, not task-solver performance. Runtime IDs and thinking settings follow Tables 8-9; the random reference is not a model. This series reports Balanced Accuracy (%). Source: https://arxiv.org/html/2606.29920v2.
Metric: Balanced Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (2026-03-05) (xHigh) | 89.4 | #63 (GPT-5.4 (2026-03-05)) |
| 2 | Gemini 3.1 Pro (Preview) (High) | 86.5 | #64 (Gemini 3.1 Pro (Preview)) |
| 3 | Claude Opus 4.7 (Max) | 85 | #51 (Claude Opus 4.7) |
| 4 | Kimi K2.6 (Thinking) | 84.3 | #113 (Kimi K2.6) |
| 5 | DeepSeek V4 Pro (Max) | 83.8 | #112 (DeepSeek V4 Pro) |
| 6 | Claude Sonnet 4.6 (Max) | 83 | #94 (Claude Sonnet 4.6) |
| 7 | DeepSeek V4 Flash (Max) | 81.3 | #155 (DeepSeek V4 Flash) |
| 8 | GLM-5.1 (Thinking) | 80.8 | #109 (GLM-5.1) |
| 9 | Qwen 3.6 Max (Preview) (Thinking) | 79.9 | |
| 10 | MiMo-V2.5-Pro (Thinking) | 78.5 | #167 (MiMo-V2.5-Pro) |
| 11 | GPT-OSS-120B (High) | 77.1 | #346 (GPT-OSS-120B) |
| 12 | Qwen 3.5 122B A10B (Thinking) | 75 | #184 (Qwen 3.5 122B A10B) |
| 13 | Qwen 3.5 27B (Thinking) | 74 | #192 (Qwen 3.5 27B) |
| 14 | GPT-OSS-20B (High) | 72.3 | #523 (GPT-OSS-20B) |
| 15 | Qwen 3.5 9B (Thinking) | 71.6 | #391 (Qwen 3.5 9B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ruverbench-agentic-coding · How It Works · Data refreshed daily, snapshot 2026-11-05.