Human Expert Alignment on FeedEval
83.2Specificity AccuracyGemma3-Inst.
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Gemma3-Inst.Model Scale=3B-scale, Fine-tuning Status=fine-tuned2026.01 | 83.2 | 89.3 | 85.3 | 89.5 | 82.2 | 70.2 | |
| Llama3-3B-Inst.Model Scale=3B-scale, Fine-tuning Status=fine-tuned2026.01 | 82 | 88 | 86.4 | 91.2 | 83.5 | 70.9 | |
| Phi-3-MiniModel Scale=3B-scale, Fine-tuning Status=fine-tuned2026.01 | 81.1 | 86 | 87.1 | 92 | 82 | 70 | |
| Qwen2-3B-Inst.Model Scale=3B-scale, Fine-tuning Status=fine-tuned2026.01 | 80.7 | 87 | 75.5 | 82.4 | 83.3 | 70.3 | |
| Gemini-2.5-ProFine-tuning Status=frozen2026.01 | 75.7 | 84.5 | 55.6 | 62.2 | 61.3 | 41.7 | |
| GPT-5.1Fine-tuning Status=frozen2026.01 | 72.9 | 83.3 | 58.4 | 69.7 | 64 | 44.5 |