Long-horizon agentic tasks on HLE Our Settings
44.4Pass@1AgentSwing
Evaluation results
| Method | Links | |
|---|---|---|
| AgentSwingBase Model=DeepSeek-v3.22026.03 | 44.4 | |
| DeepSeek-v3.2Context Management=Summary2026.03 | 43.5 | |
| DeepSeek-v3.2Context Management=Discard-All2026.03 | 42 | |
| DeepSeek-v3.2Context Management=Baseline (w/o CM)2026.03 | 40.2 | |
| DeepSeek-v3.2Context Management=Keep-Last-N2026.03 | 39.6 | |
| AgentSwingBase Model=GPT-OSS-120B2026.03 | 35.1 | |
| GPT-OSS-120BContext Management=Summary2026.03 | 34.4 | |
| GPT-OSS-120BContext Management=Discard-All2026.03 | 34.2 | |
| GPT-OSS-120BContext Management=Keep-Last-N2026.03 | 34.1 | |
| GPT-OSS-120BContext Management=Baseline (w/o CM)2026.03 | 33.2 | |
| AgentSwingBase Model=Tongyi-DR-30B-A3B2026.03 | 33.1 | |
| Tongyi-DR-30B-A3BContext Management=Discard-All2026.03 | 32.7 | |
| Tongyi-DR-30B-A3BContext Management=Keep-Last-N2026.03 | 32.2 | |
| Tongyi-DR-30B-A3BContext Management=Summary2026.03 | 32 | |
| Tongyi-DR-30B-A3BContext Management=Baseline (w/o CM)2026.03 | 31.7 |