Qwen3-Coder-30B-A3B-Instruct

Alibaba · 30.5B (active 3B) · MoE · ctx 128K · Apache 2.0 · HuggingFace

Qwen3-Coder-30B-A3B-Instruct (Alibaba) is a 30.5B parameters MoE (3B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 18.38 GB VRAM for Q4 (weights 16.78 GB + KV 0.6 GB + 1 GB overhead) and 62.6 GB for FP16. About 22 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 1697.0 tok/s (estimated); MI300X 192GB ≈ 1124.2 tok/s (estimated); H200 141GB SXM5 ≈ 1018.2 tok/s (estimated); H100 80GB SXM5 ≈ 710.6 tok/s (estimated); Gaudi 3 128GB ≈ 694.9 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

85 hardware configs — sorted by tok/s. Default fit context 8K.