Interruption Handling on Full-Duplex-Bench
4.59GPT-4o ScoreQwen-2.5-Omni
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen-2.5-Omni2026.01 | 4.59 | — | 2.74 | |
| FLAIRlatent reasoning=true2026.03 | 4.22 | 92 | 0.36 | |
| PersonaPlex2026.01 | 4.21 | 1 | 0.4 | |
| Vanilla Moshitraining=fine-tuned2026.04 | 4.19 | 0.95 | 0.52 | |
| FLAIRlatent reasoning=false2026.03 | 4.08 | 89 | 0.35 | |
| MoshiRAG2026.04 | 3.75 | 0.85 | 1.02 | |
| Freeze-Omni2026.03 | 3.62 | 86.7 | 1.41 | |
| Freeze-Omni2026.04 | 3.62 | 0.87 | 1.41 | |
| Freeze-Omni2026.01 | 3.615 | 0.867 | 1.409 | |
| Gemini Live2026.03 | 3.38 | 89.1 | 1.18 | |
| Gemini2026.04 | 3.38 | 0.89 | 1.18 | |
| Gemini2026.01 | 3.376 | 0.891 | 1.183 | |
| Moshi2026.03 | 0.77 | 100 | 0.26 | |
| Vanilla Moshi2026.04 | 0.77 | 1 | 0.26 | |
| Moshi2026.01 | 0.765 | 1 | 0.257 | |
| dGSLM2026.01 | 0.201 | 0.917 | 2.531 | |
| dGSLM2026.03 | 0.2 | 91.7 | 2.53 | |
| dGSLM2026.04 | 0.2 | 0.92 | 2.53 |