Multistep Reasoning on SpokenMQA
81.5AccuracyBaseline (Thinking)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Baseline (Thinking)Model Size=4B, Backbone=Qwen3, Hardware=A1002025.12 | 81.5 | 227.16 | 227.15 | |
| AsyncReasoningModel Size=4B, Backbone=Qwen3, Hardware=A1002025.12 | 81 | 0.84 | 0.8 | |
| Baseline (No think)Model Size=4B, Backbone=Qwen3, Hardware=A1002025.12 | 80.6 | 0.8 | 0.7 | |
| Baseline (Thinking)Model Size=0.6B, Backbone=Qwen3, Hardware=A1002025.12 | 70.4 | 201.53 | 201.52 | |
| AsyncReasoningModel Size=0.6B, Backbone=Qwen3, Hardware=A1002025.12 | 65.7 | 2.56 | 1.37 | |
| Baseline (No think)Model Size=0.6B, Backbone=Qwen3, Hardware=A1002025.12 | 56.8 | 0.72 | 0.67 |