Long-horizon agent performance on WideSearch
72.7Overall ScoreKimi-K2.5-1T
Evaluation results
| Method | Links | |
|---|---|---|
| Kimi-K2.5-1TCM=w/o2026.05 | 72.7 | |
| SAMBackbone=GLM-4.72026.05 | 69.2 | |
| SAMBackbone=Qwen3.5-35B-A3B2026.05 | 69.1 | |
| GLM-4.7CM=summary2026.05 | 68.3 | |
| GLM-4.7CM=recent-k2026.05 | 67.1 | |
| Qwen3.5-35B-A3BCM=summary2026.05 | 66.8 | |
| GLM-4.7CM=discard-tool2026.05 | 66.3 | |
| Qwen3.5-35B-A3BCM=w/o2026.05 | 65.6 | |
| GLM-4.7CM=w/o2026.05 | 65.4 | |
| Qwen3.5-35B-A3BCM=recent-k2026.05 | 65.3 | |
| Qwen3.5-35B-A3BCM=discard-tool2026.05 | 64.7 | |
| AgentFold-30B-A3BCategory=Agent systems2026.05 | 62.1 | |
| OpenAI-o3CM=w/o2026.05 | 52.6 |