Tongyi DeepResearch 30B A3B

Tongyi DeepResearch 30B A3B, developed by Alibaba's Tongyi Lab, is an agentic large language model optimized for long-horizon, deep information-seeking tasks. It exhibits strong reliability, successfully providing usable responses 89% of the time. However, its speed performance tends to be slower, ranking in the 14th percentile. The model offers competitive pricing, placing in the 43rd percentile. In terms of performance across benchmarks, Tongyi DeepResearch demonstrates exceptional proficiency in Coding, achieving 96.8% accuracy, and strong capabilities in General Knowledge (99.0%) and Email Classification (99.0%). Its Reasoning abilities are also robust at 93.6% accuracy. A notable weakness is its performance in Ethics, where it scored only 57.3% accuracy, and Mathematics (68.1%). Hallucination control is moderate at 80.0% accuracy. The model's design, featuring a 30 billion total parameter count with 3 billion active per token, and its use of advanced techniques like a synthetic data pipeline and Group Relative Policy Optimization, position it well for complex agentic search and multi-step problem-solving, despite its slower response times.