Qwen: Qwen3 30B A3B Thinking 2507

Qwen3-30B-A3B-Thinking-2507 is a cutting-edge 30B parameter Mixture-of-Experts (MoE) reasoning model, specifically engineered for intricate tasks demanding extensive multi-step thought processes. This model is uniquely designed for a “thinking mode,” where its internal reasoning traces are kept separate from the final answers, allowing for clearer and more structured problem-solving. This version significantly outperforms earlier Qwen3-30B releases across critical benchmarks, including logical reasoning, mathematics, science, coding, and multilingual capabilities. It also features enhanced instruction following, superior tool use, and stronger alignment with human preferences. With a context window of 32K tokens and a max output of 4K tokens, it offers high reasoning efficiency and extended output budgets. Priced at $0.05/$0.34 per 1M tokens (input/output) and available in the FREE access tier, it is best suited for advanced research, competitive problem solving, and sophisticated agentic applications requiring structured long-context reasoning.