Knowledge & Instruction Following on GPQA Diamond (Score)
80.3ScoreQwen3.5-4B (released)
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3.5-4B (released)Model Size=4B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.06 | 80.3 | |
| Qwen3.5-9B (released)Model Size=9B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.06 | 80.3 | |
| Qwen3.5-9BParams=9B, Sampling mode=AR2026.06 | 80.3 | |
| Mercury-2Params=Commercial, Sampling mode=Diffusion2026.06 | 73 | |
| Qwen3.5-9B + AR-SFTModel Size=9B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.06 | 71.21 | |
| FLARE-9BModel Size=9B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.06 | 71.21 | |
| FLARE-9BParams=9B, Sampling mode=AR-Trust2026.06 | 71.21 | |
| LLaDA-2.1-flashParams=100B-A5B, Sampling mode=Diffusion2026.06 | 66.67 | |
| FLARE-9BParams=9B, Sampling mode=Diffusion-Trust2026.06 | 64.65 | |
| FLARE-4BModel Size=4B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.06 | 63.64 | |
| Qwen3.5-2B (released)Model Size=2B, Training Protocol=Released, Decoding Strategy=Native AR decoding2026.06 | 62.12 | |
| LLaDA-2.0-flashParams=100B-A5B, Sampling mode=Diffusion2026.06 | 61.98 | |
| Qwen3.5-4B + AR-SFTModel Size=4B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.06 | 61.11 | |
| LLaDA-2.1-miniParams=16B-A1B, Sampling mode=Diffusion2026.06 | 48.36 | |
| Qwen3.5-2B + AR-SFTModel Size=2B, Training Protocol=AR-SFT, Decoding Strategy=Native AR decoding2026.06 | 47.98 | |
| LLaDA-2.0-miniParams=16B-A1B, Sampling mode=Diffusion2026.06 | 47.98 | |
| FLARE-2BModel Size=2B, Training Protocol=FLARE, Decoding Strategy=AR-Trust sampling2026.06 | 37.37 | |
| SDAR 30B-A3BParams=30B-A3B, Sampling mode=Diffusion2026.06 | 36.7 |