RoadmapBench - Web and RPC: leaderboard
Metric: Resolved rate (%) on the 17 web and RPC 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 April 2027. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.7 (xHigh) | 41.2 |
| 2 | Claude Opus 4.6 (Thinking) | 29.4 |
| 3 | Gemini 3.1 Pro (Preview) (High) | 23.5 |
| 4 | GPT-5.4 (High) | 17.6 |
| 5 | DeepSeek V4 Pro (High) | 17.6 |
| 6 | Qwen 3.5 397B A17B | 11.8 |
| 7 | GLM-5.1 | 11.8 |
| 8 | MiMo-V2.5-Pro | 11.8 |
| 9 | Kimi K2.5 | 5.9 |
| 10 | Kimi K2.6 | 5.9 |
| 11 | Qwen 3.6 Plus | 5.9 |
| 12 | Seed 2.0 Pro (High) | 5.9 |
| 13 | MiniMax-M2.7 | 0 |
Interactive version: theaggregate.ai/benchmark?slug=roadmapbench-web-and-rpc · How It Works · Data refreshed daily, snapshot 2026-10-07.