Agentic Task Performance on AssistantBench (test)
65.8Easy AccuracyLearning to Share
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Learning to ShareShared Memory=true, Model=Qwen3 (32B)2026.02 | 65.8 | 25.2 | 14.4 | 20.3 | 1,479 | |
| Learning to ShareShared Memory=true, Model=GPT-5.12026.02 | 61 | 35.1 | 20 | 26.7 | 882 | |
| M1-ParallelShared Memory=false, Model=GPT-5.12026.02 | 57.3 | 35.2 | 16 | 24 | 1,389 | |
| MagenticOneShared Memory=false, Model=Qwen3 (32B)2026.02 | 53.1 | 17.4 | 8.5 | 13.4 | 1,084 | |
| M1-ParallelShared Memory=false, Model=Qwen3 (32B)2026.02 | 52.6 | 21.4 | 8.5 | 14.7 | 2,239 | |
| MagenticOneShared Memory=false, Model=GPT-5.12026.02 | 43.6 | 31.5 | 15.3 | 21.7 | 724 |