ORAgentBench - Hard: leaderboard
Metric: Pass rate (%) on the 34 hard tasks: a task passes when the submission is feasible and its normalized quality reaches 0.4; end-to-end operations research tasks packaged with multi-file operational data, configuration and a submission schema; the agent writes and runs solver code in an isolated Harbor environment (45-minute interaction budget, 5-minute solve limit) and hidden validators score schema and hard-constraint feasibility and normalized objective quality; higher is better. Source: arxiv.org. Saturation forecast: Around January 2028. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (High) | 20.59 |
| 2 | GPT-5.3 Codex (High) | 20.59 |
| 3 | Claude Sonnet 4.6 | 14.71 |
| 4 | Claude Opus 4.6 | 14.71 |
| 5 | DeepSeek V4 Pro | 13.73 |
| 6 | Kimi K2.6 | 11.76 |
| 7 | GLM-5.1 | 8.82 |
| 8 | DeepSeek V4 Flash | 5.88 |
| 9 | Qwen 3.6 Plus | 5.88 |
| 10 | GLM-5 | 5.88 |
| 11 | GPT-5.4 Mini (High) | 5.88 |
| 12 | MiniMax-M2.7 | 2.94 |
| 13 | MiMo-V2.5-Pro | 2.94 |
| 14 | Qwen 3.5 Plus | 2.94 |
Interactive version: theaggregate.ai/benchmark?slug=oragentbench-hard · How It Works · Data refreshed daily, snapshot 2026-09-29.