Qwen 3 30B

Qwen3-30B-Instruct is an instruction-tuned MoE model with 30.5 billion total parameters and 3.3 billion activated through sparse expert routing. Tuned for instruction following and user queries, it performs well across reasoning, mathematics, coding, and knowledge tasks while keeping multilingual capability across 100+ languages. It supports a 131K context window for long documents, extended reasoning, and large code files. The mixture-of-experts design balances cost and capability, activating only 3.3 billion of 30.5 billion parameters per token. Tool calling and agentic capabilities support more complex workflows. This is an FP8 quantized build that trims the memory footprint for GPU deployment without degrading instruction-following quality. It draws on the broad pretraining of the Qwen3 family, with strong results on mathematical reasoning, software development, and creative writing. Open-weight under Apache 2.0.