RoadmapBench - ORM and Validation: leaderboard
Metric: Resolved rate (%) on the 25 ORM and validation tasks (every subtask test suite passes), 115 long-horizon tasks implementing a real release upgrade from a roadmap of subtasks, one trial per model, 2-hour wall-clock budget, extended thinking on; each row names its agent scaffold (OpenHands, the primary scaffold, or Terminus 2, the ablation); higher is better. Source: arxiv.org. Saturation forecast: Around March 2028. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 (Thinking) | 32 |
| 2 | Claude Opus 4.7 (xHigh) | 32 |
| 3 | GLM-5.1 | 28 |
| 4 | Gemini 3.1 Pro (Preview) (High) | 24 |
| 5 | DeepSeek V4 Pro (High) | 24 |
| 6 | Kimi K2.6 | 20 |
| 7 | GPT-5.4 (High) | 20 |
| 8 | Kimi K2.5 | 12 |
| 9 | Qwen 3.5 397B A17B | 12 |
| 10 | Qwen 3.6 Plus | 12 |
| 11 | MiMo-V2.5-Pro | 12 |
| 12 | MiniMax-M2.7 | 8 |
| 13 | Seed 2.0 Pro (High) | 8 |
Interactive version: theaggregate.ai/benchmark?slug=roadmapbench-orm-and-validation · How It Works · Data refreshed daily, snapshot 2026-10-07.