According to Qdaily News, Alibaba’s PAI team has released and open-sourced a small intelligent agent language model, AgenticQwen (including two versions: 8B and 30B-a3b), designed specifically for industrial-grade tool invocation. This model series is trained via an innovative “dual data flywheel” reinforcement learning framework, which significantly reduces inference costs while delivering agent capabilities comparable to those of models with nearly a trillion parameters. Evaluation shows that AgenticQwen-8B achieves an average score of 47.4 on real-world tool environment benchmarks (such as tau-2 and bfcl-v4), far exceeding the base version Qwen3-8B (23.8) and approaching Qwen3-235B (52.0). The model has now been deployed in internal production systems similar to Manus, substantially narrowing the gap with the 235B large model.