Pedagogical Knowledge Evaluation on CDPK (Cross-Domain Pedagogical Knowledge) Benchmark
96.52Replicated AccuracyEDUQWEN 32B-SFT-RL2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| EDUQWEN 32B-SFT-RL2Inference-time compute=RL 2ND ROUND2026.04 | 96.52 | — | |
| EDUQWEN 32B-SFTInference-time compute=SFT2026.04 | 96.2 | — | |
| EDUQWEN 32B-RL1Inference-time compute=RL 1ST ROUND2026.04 | 94.13 | — | |
| QWEN3-235B-A22B-THINKINGInference-time compute=RANDOM FEW-SHOT2026.04 | 86.63 | 85.65 | |
| QWEN3-32B BASE MODELInference-time compute=RANDOM FEW-SHOT2026.04 | 83.37 | 82.42 | |
| QWEN2.5-72B-INSTRUCTInference-time compute=RANDOM FEW-SHOT2026.04 | 78.6 | 75.42 | |
| GEMINI-3PROInference-time compute=BENCHMARK LEADER2026.04 | — | 90.55 |