Coding Agent Leaderboard - Shellbench (OpenClaw): leaderboard
Red Hat's Coding Agent Leaderboard run of Shellbench — 115 shell and tool-use tasks from ShellBench's task set, with the OpenClaw agent scaffold in Harbor sandboxes; percentage of tasks solved. The two GLM-5.2 runs list 109 tasks rather than 115.
Metric: Score (%). Source: huggingface.co. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.2 FP8 | 18.7 | #388 |
| 2 | GLM-5.2 | 18.3 | #138 |
| 3 | GPT-OSS-120B | 0 | #700 |
Interactive version: theaggregate.ai/benchmark?slug=coding-agent-leaderboard-shellbench-openclaw · How It Works · Data refreshed daily, snapshot 2026-11-13.