DataClawBench - Hard: leaderboard
Metric: Final-answer accuracy (%) on the 75 hard tasks, OpenClaw agent (Python execution, file reading, web fetch and web search tools) in an isolated container with the read-only data environment, vendor-default sampling and reasoning, 1200 s per task, GLM-5 judge against the gold answer with sub-questions scored separately; higher is better. Source: arxiv.org. Saturation forecast: Around October 2028. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 (High) | 44.8 |
| 2 | MiniMax-M2.7 | 19.3 |
| 3 | Qwen 3.5 Plus | 16.2 |
| 4 | Kimi K2.5 | 14 |
| 5 | GLM-5 | 12.6 |
| 6 | Gemini 3.1 Pro (Preview) (High) | 12.3 |
| 7 | GPT-5.4 (Non-reasoning) | 12 |
| 8 | DeepSeek V3.2 (Thinking) | 8.9 |
Interactive version: theaggregate.ai/benchmark?slug=dataclawbench-hard · How It Works · Data refreshed daily, snapshot 2026-10-07.