| Benchmark | Model | Rank | Results |
|---|---|---|---|
| arithmetic-reasoning-on-gsm8k | Trust-Region behavior Blending (Qwen3-0.6B-Base ← Qwen3-4B) | – | Accuracy: 70.1 |
| mathematical-reasoning-on-aime-2025 | Trust-Region behavior Blending (Qwen3-1.7B-Base ← Qwen3-8B) | – | Accuracy: 6.9 |
| mathematical-reasoning-on-aime24 | Trust-Region behavior Blending (Qwen3-1.7B-Base ← Qwen3-8B) | – | Acc: 10.2 |
| mathematical-reasoning-on-math-500 | Trust-Region behavior Blending (Qwen3-1.7B-Base ← Qwen3-8B) | – | Accuracy: 69.7 |