Sentiment Classification on SST-2 (Nec, Suf, Shuf Metrics)
37.1Nec ScoreMiniMax-M2.5
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MiniMax-M2.5N=427–496, Pattern=Genuine2026.03 | 37.1 | 60.7 | 38.1 | |
| GLM-5N=392, Pattern=Decorative2026.03 | 17.8 | 82.3 | 15.4 | |
| Kimi-K2.5N=457–500, Pattern=Decorative2026.03 | 16.5 | 79.5 | 18.5 | |
| Claude Opus 4.6-RN=486–499, Pattern=Decorative2026.03 | 14.8 | 91.4 | 22.4 | |
| DeepSeek-V3.2N=500, Pattern=Decorative2026.03 | 10.8 | 96.7 | 16.4 | |
| Qwen3.5-397BN=98, Pattern=Decorative2026.03 | 8.2 | 96.9 | 5.1 | |
| Small models (0.8–8B avg)N=100 ea., Pattern=Task-split2026.03 | 8 | 76 | — | |
| Nemotron-UltraN=306–498, Pattern=Ctx Dep†2026.03 | 6.4 | 68 | 18.1 | |
| GPT-OSS-120BN=425–496, Pattern=Decorative2026.03 | 0.3 | 98.9 | 5.2 | |
| GPT-5.4N=376–500, Pattern=Decorative2026.03 | 0.1 | 98.2 | 0.1 | |
| Qwen3.5-122BN=343, Pattern=Decorative2026.03 | 0 | 98.5 | 0.3 |