DisasterBench - Chain-of-Thought: leaderboard
Metric: Exact-match accuracy (%) over the 233 DisasterBench planning tasks (35 single-tool, 166 chain and 32 branching workflows over 26 disaster-response tools): a plan counts only when every step matches the ground truth in tool identity, parameter bindings and inter-step dependencies; planning method: intermediate reasoning before the final plan (temperature 0); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 72.53 |
| 2 | Qwen 3 Max (Thinking) | 66.52 |
| 3 | Gemma 4 31B | 63.09 |
| 4 | DeepSeek R1 | 62.66 |
| 5 | Qwen 3 Max | 62.66 |
| 6 | GPT-5.4 | 62.23 |
| 7 | Qwen 3.5 27B | 62.23 |
| 8 | Qwen 3.5 9B | 61.37 |
| 9 | Llama 3.3 70B | 58.8 |
| 10 | DeepSeek V3.2 | 36.05 |
| 11 | Ministral 3 14B | 33.48 |
| 12 | Llama 3.1 8B | 26.18 |
Interactive version: theaggregate.ai/benchmark?slug=disasterbench-chain-of-thought · How It Works · Data refreshed daily, snapshot 2026-10-07.