Language Understanding on MMLU (Score and Relative Gain)
90.8MMLU ScoreDeepSeek V3.2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepSeek V3.2SLM=DeepSeek V3.22026.05 | 90.8 | — | |
| +PACESLM=Qwen3.5-9B2026.05 | 88.9 | 8.6 | |
| +PACESLM=Ministral-3-14B2026.05 | 85.4 | 6.7 | |
| +CESLM=Ministral-3-14B2026.05 | 83.6 | 4.5 | |
| Gödel AgentSLM=Qwen3.5-9B2026.05 | 83.5 | 2.1 | |
| +CESLM=Qwen3.5-9B2026.05 | 83.5 | 2.1 | |
| MIPROv2SLM=Ministral-3-14B2026.05 | 83.3 | 4.1 | |
| GEPASLM=Qwen3.5-9B2026.05 | 83.2 | 1.7 | |
| +PESLM=Qwen3.5-9B2026.05 | 83.2 | 1.7 | |
| +PESLM=Ministral-3-14B2026.05 | 83.2 | 4.4 | |
| GEPASLM=Ministral-3-14B2026.05 | 82.5 | 3.1 | |
| +PACESLM=Qwen3-4B2026.05 | 81.8 | 6.1 | |
| VanillaSLM=Qwen3.5-9B2026.05 | 81.8 | — | |
| ACESLM=Ministral-3-14B2026.05 | 81.8 | 2.2 | |
| ACESLM=Qwen3.5-9B2026.05 | 80.5 | -1.5 | |
| ACESLM=Qwen3-4B2026.05 | 80 | 3.7 | |
| VanillaSLM=Ministral-3-14B2026.05 | 80 | — | |
| Gödel AgentSLM=Qwen3-4B2026.05 | 79.7 | 3.3 | |
| +CESLM=Qwen3-4B2026.05 | 79.7 | 3.3 | |
| MIPROv2SLM=Qwen3.5-9B2026.05 | 79.7 | -2.5 | |
| Gödel AgentSLM=Ministral-3-14B2026.05 | 79.7 | 0 | |
| +PESLM=Qwen3-4B2026.05 | 78.8 | 2.2 | |
| VanillaSLM=Qwen3-4B2026.05 | 77.1 | — | |
| MIPROv2SLM=Qwen3-4B2026.05 | 77.1 | 0 | |
| GEPASLM=Qwen3-4B2026.05 | 77.1 | 0 |