Multi-task Language Understanding on MMLU Pro (test)
92.8Physics ScoreExpeL + Janus
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| ExpeL + JanusBackbone LLM=Qwen3-8B2026.06 | 92.8 | — | — | — | — | — | — | — | 71.6 | |
| ExpeLBackbone LLM=Qwen3-8B2026.06 | 90.4 | — | — | — | — | — | — | — | 68.8 | |
| DC-RS + JanusBackbone LLM=Qwen3-8B2026.06 | 89.2 | — | — | — | — | — | — | — | 68 | |
| DC-RS + JanusBackbone LLM=DeepSeek-V4-Flash2026.06 | 89.2 | — | — | — | — | — | — | — | 74.4 | |
| ExpeL + JanusBackbone LLM=DeepSeek-V4-Flash2026.06 | 88.4 | — | — | — | — | — | — | — | 75.6 | |
| ExpRAGBackbone LLM=Qwen3-8B2026.06 | 87.2 | — | — | — | — | — | — | — | 66.2 | |
| ExpeLBackbone LLM=DeepSeek-V4-Flash2026.06 | 85.2 | — | — | — | — | — | — | — | 73.2 | |
| Memory-freeBackbone LLM=Qwen3-8B2026.06 | 84 | — | — | — | — | — | — | — | 61.2 | |
| DC-RSBackbone LLM=Qwen3-8B2026.06 | 84 | — | — | — | — | — | — | — | 64.4 | |
| DC-RSBackbone LLM=DeepSeek-V4-Flash2026.06 | 83.2 | — | — | — | — | — | — | — | 66.4 | |
| ExpRAGBackbone LLM=DeepSeek-V4-Flash2026.06 | 82.4 | — | — | — | — | — | — | — | 70 | |
| Memory-freeBackbone LLM=DeepSeek-V4-Flash2026.06 | 78.2 | — | — | — | — | — | — | — | 68.4 | |
| RL-ExtraModel Scale=Qwen2.5-14B-Instruct, Training Steps=2502026.04 | 69.8 | 61.1 | 35.2 | 83.4 | 49.1 | 71 | 65.4 | 62.1 | — | |
| NExtModel Scale=Qwen2.5-14B-Instruct, Training Steps=2502026.04 | 69.8 | 65.1 | 32.1 | 84.7 | 47.9 | 71.1 | 67 | 62.5 | — | |
| GRPOModel Scale=Qwen2.5-14B-Instruct, Training Steps=4002026.04 | 69.3 | 63.6 | 33.3 | 83.8 | 47.7 | 71.5 | 68.8 | 62.6 | — | |
| AlphaRLModel Scale=Qwen2.5-14B-Instruct, Training Steps=2502026.04 | 67.5 | 62.8 | 32.1 | 85.2 | 45.1 | 73.6 | 66.9 | 61.9 | — | |
| BackboneModel Scale=Qwen2.5-14B-Instruct, Training Steps=02026.04 | 64.7 | 64.2 | 32.3 | 83.1 | 44.2 | 72.4 | 64.9 | 60.8 | — | |
| GRPOModel Scale=Qwen2.5-7B-Instruct, Training Steps=4002026.04 | 59.4 | 49.4 | 31.1 | 70.7 | 41 | 61.6 | 54.4 | 52.5 | — | |
| NExtModel Scale=Qwen2.5-7B-Instruct, Training Steps=2502026.04 | 59.2 | 48.3 | 32.4 | 71.5 | 41.5 | 63.9 | 55.4 | 53.2 | — | |
| AlphaRLModel Scale=Qwen2.5-7B-Instruct, Training Steps=2502026.04 | 58.7 | 47.4 | 31.8 | 70 | 39.5 | 61.8 | 54.4 | 51.9 | — | |
| RL-ExtraModel Scale=Qwen2.5-7B-Instruct, Training Steps=2502026.04 | 58.7 | 51.1 | 31.3 | 70.6 | 41 | 64.2 | 53.9 | 53 | — | |
| BackboneModel Scale=Qwen2.5-7B-Instruct, Training Steps=02026.04 | 53.6 | 46.6 | 31 | 70.3 | 36.3 | 61.3 | 52.2 | 50.2 | — | |
| AlphaRLModel Scale=Qwen2.5-3B-Instruct, Training Steps=2502026.04 | 41.9 | 32.4 | 19.6 | 60.8 | 33 | 52.8 | 37.9 | 39.8 | — | |
| GRPOModel Scale=Qwen2.5-3B-Instruct, Training Steps=4002026.04 | 41.7 | 34.1 | 20.5 | 60.6 | 35.5 | 54.2 | 36.8 | 40.5 | — | |
| RL-ExtraModel Scale=Qwen2.5-3B-Instruct, Training Steps=2502026.04 | 41.2 | 36.9 | 18.5 | 59.9 | 35.5 | 54.2 | 35.6 | 40.3 | — | |
| NExtModel Scale=Qwen2.5-3B-Instruct, Training Steps=2502026.04 | 40.2 | 35.3 | 20.3 | 60.8 | 35.8 | 54.2 | 36.5 | 40.4 | — | |
| BackboneModel Scale=Qwen2.5-3B-Instruct, Training Steps=02026.04 | 36 | 32.4 | 18.3 | 55 | 35.7 | 51.9 | 37.5 | 38.1 | — | |
| NExtModel Scale=Qwen2.5-1.5B-Instruct, Training Steps=2502026.04 | 22 | 17.6 | 18.3 | 45.2 | 20.9 | 41.1 | 26.1 | 27.3 | — | |
| GRPOModel Scale=Qwen2.5-1.5B-Instruct, Training Steps=4002026.04 | 21.2 | 18.8 | 19.4 | 44.7 | 19.8 | 39.8 | 26.4 | 27.2 | — | |
| RL-ExtraModel Scale=Qwen2.5-1.5B-Instruct, Training Steps=2502026.04 | 20.7 | 19.3 | 16.5 | 38.7 | 16.9 | 38.8 | 27.2 | 25.4 | — | |
| AlphaRLModel Scale=Qwen2.5-1.5B-Instruct, Training Steps=2502026.04 | 18 | 16.1 | 17 | 40.6 | 20.9 | 40.3 | 26.7 | 25.7 | — | |
| BackboneModel Scale=Qwen2.5-1.5B-Instruct, Training Steps=02026.04 | 16.1 | 16.2 | 8.5 | 29.1 | 14.5 | 21.7 | 18 | 17.7 | — |