Multi-round coreference resolution on MRCR
81.62-Needle ScoreDeepSeek-V3.2-Thinking
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek-V3.2-Thinking2026.05 | 81.6 | 60.32 | 71.01 | |
| Qwen3-30B-A3B-Thinking + ACC2026.05 | 76.9 | 59.57 | 68.28 | |
| Qwen3-235B-A22B-Thinking2026.05 | 74.98 | 59.96 | 67.51 | |
| GPT-5.1-Thinking2026.05 | 73.75 | 47.93 | 60.91 | |
| GLM-4.6-Thinking2026.05 | 73.63 | 55.33 | 64.53 | |
| Kimi-K2-Thinking2026.05 | 68.01 | 47.96 | 58.05 | |
| Qwen3-30B-A3B-Thinking2026.05 | 61.84 | 38.41 | 50.19 | |
| GPT-OSS-120B2026.05 | 46.72 | 29.16 | 38 |