Qwen3 Technical Report

Team, 2025, Qwen3 Technical Report (arXiv preprint arXiv:2505.09388)

Subscriber deck — free during early access

1 cards in “4.2 Reasoning RL” · whole deck

4.2 Reasoning RL

What did Qwen3’s reasoning RL stage involve?
3,995 query–verifier pairs, unseen in the cold start, learnable but as hard as possible, trained with GRPO using large batches, many rollouts per query and off-policy reuse, while keeping the model’s entropy steady or slowly rising. AIME’24 for the 235B model rose from 70.1 to 85.1 in 170 steps.
- Team, 2025, Qwen3 Technical Report (arXiv preprint arXiv:2505.09388), §4.2

Source

Qwen Team. 2025. Qwen3 Technical Report. In arXiv preprint arXiv:2505.09388. https://arxiv.org/abs/2505.09388