Mathematical Reasoning on GSM8K (Accuracy Ref vs Resp)
70.8Accuracy (Ref)MoshiRAG_GPT-4.o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MoshiRAG_GPT-4.oreference generator=GPT-4.o2026.04 | 70.8 | 43.2 | |
| MoshiRAG2026.04 | 66.2 | 33.9 | |
| STITCH-S2026.04 | 56.7 | — | |
| GLM-4-Voice2026.04 | 29 | — | |
| MoshiRAGsumm.summarized reference=true2026.04 | — | 51.2 | |
| Vanilla Moshi2026.04 | — | 2.1 |