Reasoning on DART 3
85.5AccuracyDeepSeek-R1
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-R1Setting=ARM only2026.03 | 85.5 | 2,076 | |
| Qwen3-1.7BSetting=ARM only2026.03 | 83 | 2,112 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=642026.03 | 57 | 5 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=2562026.03 | 50 | 5 | |
| Latent-DARM (LLaDA-8B → Llama-3.2-3B)Setting=latent-space, Plan tokens=1282026.03 | 43 | 5 | |
| LLaDA-8B → Llama-3.2-3BSetting=text-space2026.03 | 35.5 | 16 | |
| Llama-3.2-3BSetting=ARM only2026.03 | 28 | 22 |