Multi-turn conversation performance on Math
94.5Avg PerformanceFull
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FullModel=GPT-5.2, Setting=Ideal instructions2026.02 | 94.5 | 89.4 | |
| FullModel=DeepSeek-v3.2-Thinking, Setting=Ideal instructions2026.02 | 94 | 81.6 | |
| FullModel=GPT-4o-mini, Setting=Ideal instructions2026.02 | 87.2 | 70.9 | |
| Experience-Driven MediatorModel=DeepSeek-v3.2-Thinking, Setting=Ambiguous user inputs with Mediator2026.02 | 86.3 | 67.3 | |
| Experience-Driven MediatorModel=GPT-5.2, Setting=Ambiguous user inputs with Mediator2026.02 | 80.6 | 62 | |
| ShardedModel=DeepSeek-v3.2-Thinking, Setting=Ambiguous user inputs2026.02 | 78.8 | 56.3 | |
| Experience-Driven MediatorModel=GPT-4o-mini, Setting=Ambiguous user inputs with Mediator2026.02 | 77.7 | 70.4 | |
| ShardedModel=GPT-5.2, Setting=Ambiguous user inputs2026.02 | 69.6 | 48.6 | |
| ShardedModel=GPT-4o-mini, Setting=Ambiguous user inputs2026.02 | 64.9 | 45.6 |