Dialogue State Tracking on SGD
80.68Joint GAReacTOD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ReacTODModel=Claude-Opus-4.6, predicts domains end-to-end=true, impractical for production latency=true2026.05 | 80.68 | — | |
| ReacTODModel=Qwen3-32B, predicts domains end-to-end=true2026.05 | 64.09 | — | |
| ReacTODModel=gpt-oss-20B, predicts domains end-to-end=true2026.05 | 62.92 | — | |
| ReacTODModel=Qwen3-8B, predicts domains end-to-end=true2026.05 | 57.31 | — | |
| ReacTODModel=Gemma3-12B, predicts domains end-to-end=true2026.05 | 55.58 | — | |
| SRPModel=Claude-Opus-4.6, gold domain labels=true, per-domain isolated sessions=true, reproduced=true2026.05 | 45.2 | — | |
| Seq2Seq-DUBackbone=BERT-Base, Uncased2020.11 | 30.1 | 91 | |
| FastSGT2020.11 | 29.2 | 90.3 | |
| SGD-baseline2020.11 | 25.4 | 90.6 |