Reasoning on DART 4
75AccuracyDeepSeek-R1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-R1Setting=ARM only2026.03 | 75 | 2,418 | |
| Qwen3-1.7BSetting=ARM only2026.03 | 73 | 2,550 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=642026.03 | 63 | 5 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=2562026.03 | 62 | 5 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=1282026.03 | 49 | 5 | |
| Llama-3.2-3BSetting=ARM only2026.03 | 34.5 | 28 | |
| LLaDA-8B → Llama-3.2-3BSetting=text-space2026.03 | 30 | 12 |