Opinion Alignment on ANES
62.33Mean AccuracyGRPO
Evaluation results
| Method | Links | |
|---|---|---|
| GRPOBase model=Magistral 24B2026.03 | 62.33 | |
| SFT+GRPOBase model=Magistral 24B2026.03 | 62.33 | |
| SFT+GRPOBase model=Llama 3.1 8B2026.03 | 59.21 | |
| SFTBase model=Llama 3.1 8B2026.03 | 58.82 | |
| SFT+GRPOBase model=Qwen3 8B2026.03 | 55.14 | |
| GRPOBase model=Qwen3 8B2026.03 | 52.92 | |
| SFTBase model=Magistral 24B2026.03 | 52.8 | |
| GRPOBase model=Llama 3.1 8B2026.03 | 52.35 | |
| SFTBase model=Qwen3 8B2026.03 | 47.01 | |
| iclBase model=Llama 3.1 8B2026.03 | 44.39 | |
| ORPOBase model=Llama 3.1 8B2026.03 | 44.15 | |
| iclBase model=Magistral 24B2026.03 | 42.46 | |
| iclBase model=Qwen3 8B2026.03 | 41.32 | |
| ORPOBase model=Qwen3 8B2026.03 | 34.07 | |
| randomBase model=Untrained baselines2026.03 | 33.33 | |
| ORPOBase model=Magistral 24B2026.03 | 31.32 | |
| majorityBase model=Untrained baselines2026.03 | 22.98 |