ORAgentBench - Easy: leaderboard

Metric: Pass rate (%) on the 32 easy 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 2034. 14 models tracked.

Top models

#ModelScore
1Claude Opus 4.650
2DeepSeek V4 Flash46.88
3Qwen 3.5 Plus46.88
4GPT-5.4 (High)46.88
5MiMo-V2.5-Pro43.75
6GPT-5.4 Mini (High)43.75
7Qwen 3.6 Plus40.62
8MiniMax-M2.740.62
9DeepSeek V4 Pro39.58
10GLM-5.137.5
11GLM-537.5
12Claude Sonnet 4.631.25
13Kimi K2.631.25
14GPT-5.3 Codex (High)31.25

Interactive version: theaggregate.ai/benchmark?slug=oragentbench-easy · How It Works · Data refreshed daily, snapshot 2026-09-29.