ORAgentBench - Hard Quality: leaderboard
Metric: Mean normalized objective quality (0-100) on the 34 hard tasks, zero for infeasible submissions and 0.5 at the verified reference solution; 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; printed 0-1 and x100 here; higher is better. Source: arxiv.org. Saturation forecast: Around April 2028. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (High) | 21 |
| 2 | GPT-5.3 Codex (High) | 19 |
| 3 | Claude Sonnet 4.6 | 14 |
| 4 | Claude Opus 4.6 | 14 |
| 5 | DeepSeek V4 Pro | 14 |
| 6 | Kimi K2.6 | 13 |
| 7 | GLM-5.1 | 10 |
| 8 | DeepSeek V4 Flash | 9 |
| 9 | Qwen 3.6 Plus | 7 |
| 10 | GPT-5.4 Mini (High) | 7 |
| 11 | GLM-5 | 6 |
| 12 | MiMo-V2.5-Pro | 4 |
| 13 | MiniMax-M2.7 | 3 |
| 14 | Qwen 3.5 Plus | 3 |
Interactive version: theaggregate.ai/benchmark?slug=oragentbench-hard-quality · How It Works · Data refreshed daily, snapshot 2026-09-29.