Reasoning on DART-5
61.5AccuracyDeepSeek-R1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-R1Setting=ARM only2026.03 | 61.5 | 3,068 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=642026.03 | 54 | 6 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=2562026.03 | 52.5 | 6 | |
| Qwen3-1.7BSetting=ARM only2026.03 | 49 | 3,106 | |
| Llama-3.2-3BSetting=ARM only2026.03 | 36.5 | 41 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=1282026.03 | 36 | 6 | |
| LLaDA-8B → Llama-3.2-3BSetting=text-space2026.03 | 27 | 20 |