Instruction Conflict on Oyster-II Instruction Conflict Suite
87.88Domain Constraint AdherenceQwen3-Max
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-Max2026.07 | 87.88 | 86.11 | 78.31 | 100 | 96.3 | 93.98 | 94.37 | 87.5 | 85.68 | |
| Oyster-II2026.07 | 86.36 | 70.37 | 84.14 | 98.57 | 94.44 | 89.39 | 91.25 | 91.3 | 88.23 | |
| Qwen3.5-397BModel Size=397B2026.07 | 63.64 | 80.18 | 92.77 | 100 | 96.3 | 97.59 | 87.5 | 71.59 | 90.02 | |
| Oyster-I2026.07 | 62.12 | 60.19 | 50.6 | 92.86 | 66.67 | 64.18 | 55.69 | 37.68 | 60.49 | |
| Qwen3-14BModel Size=14B2026.07 | 53.03 | 60.19 | 63.25 | 78.57 | 66.67 | 73.13 | 66.17 | 33.33 | 60.31 |