DeepPlanning
Agents · 2026-01-26
DeepPlanning benchmarks long-horizon agentic planning in multi-day travel and multi-product shopping tasks that require tool use to gather information and plans that satisfy verifiable global constraints such as time and budget. Scores report the case-level success rate of plans that satisfy every constraint.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.7-Plus (source unspecified) | 62.3 |
| Opus 4.6 | 58.9 |
| GPT-5.2 | 44.6 |
| DeepSeek-V4-Pro | 44.6 |
| Kimi K2.6 | 42.3 |
| GLM-5.1 | 34.1 |
| Opus 4.5 | 33.9 |
| Qwen3 Max Thinking (2026-01-23) | 28.7 |
| Qwen3.6 35B A3B | 25.9 |
| Qwen3.5 122B A10B | 24.1 |
| Gemma 4 31B IT | 24.0 |
| Gemini 3 Pro Preview | 23.3 |
| Qwen3.5 35B A3B | 22.8 |
| Qwen3.5-27B | 22.6 |
| Qwen3.5-9B | 18.0 |
| GPT-5 Mini | 17.9 |
| Qwen3.5-4B | 17.6 |
| Qwen3-235B-A22B | 17.1 |
| Gemma 4 26B A4B IT | 16.2 |
| GLM-5 | 14.6 |
| Qwen3 30B A3B Thinking 2507 | 4.9 |
| Qwen3-Next-80B-A3B-Thinking | 0.4 |