Language Understanding on MMLU-Pro (Accuracy and Token Counts)
88.93AccuracyVanilla MAS + MOC
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Vanilla MAS + MOCAgent Backbone=DeepSeek-V3.2-685B-Chat2026.06 | 88.93 | 3,722,630 | 759,219 | |
| Vanilla MASAgent Backbone=DeepSeek-V3.2-685B-Chat2026.06 | 88.21 | 3,178,655 | 794,530 | |
| Single LLMAgent Backbone=DeepSeek-V3.2-685B-Chat2026.06 | 66.79 | 85,260 | 1,120 |