How much VRAM to run Qwen3 Coder 30B-A3B?

About 22 GB atQ4_K_M with an 8K context — fits a RX 7900 XTX. Full breakdown below, or check your exact hardware.

Qwen3 Coder 30B-A3B VRAM by quantisation

QuantisationWeightsTotal (8K ctx)Fits on
Q4_K_M18.6 GB22.4 GBRX 7900 XTX, RTX 4090
Q5_K_M21.7 GB25.8 GBRTX 5090, Radeon AI PRO R9700
Q6_K25.0 GB29.4 GBRTX 5090, Radeon AI PRO R9700
Q8_032.5 GB37.6 GBRTX 6000 Ada, Apple Silicon 64GB unified
FP16 / BF1661.0 GB69.0 GBA100 80GB, RTX PRO 6000 Blackwell

Check your hardware

About Qwen3 Coder 30B-A3B

Qwen3 Coder 30B-A3B is the earlier Qwen3 generation's coding specialist, kept relevant by a simple formula: a mixture-of-experts design with a very small active set, tuned specifically for code generation, repository navigation and agentic coding loops, under Apache 2.0. Unlike the later hybrid-attention generations it uses conventional grouped-query attention throughout.

It earned a place as a standard local pairing for code assistants and editor integrations — capable enough for real work, light enough in active compute to feel responsive during interactive sessions.

Qwen3 Coder 30B-A3B is Alibaba's 30.5B-parameter model released in July 2025, with a 256K-token context window. It is a mixture-of-experts model: all 30.5B parameters must sit in memory, but only ~3.3B are active per token, which is what makes it fast for its size. It uses a classic dense-attention design whose KV cache grows linearly with context: its KV cache is about 0.8 GB at an 8K context, 12.9 GB at 128K, and 25.8 GB at the full 256K window (FP16 cache).

The resident-weights rule puts its quantised total just above true mainstream cards and comfortably within high-VRAM consumer territory, as the table shows, while the tiny active set delivers the responsiveness that matters most in an editor loop. Its classic dense attention means the KV cache grows linearly with context in the traditional way — and coding sessions accumulate context fast — so the calculator above is worth a look at your realistic session length, not just the default. It is a first-class citizen in the local tooling ecosystem, with well-established quantisations across formats.

For most people Q4_K_M is the sweet spot — the most popular quality/size trade-off — while Q8 is near-lossless if you have the memory. Totals above include the KV cache and a realistic framework overhead, so they are what you should expect to see in practice rather than just the download size. Weight sizes are calibrated against real GGUF files — see the methodology.

Frequently asked questions

How much VRAM does Qwen3 Coder 30B-A3B need?

At Q4_K_M with an 8K context, Qwen3 Coder 30B-A3B needs about 22 GB (weights 19 GB + KV cache + overhead). The smallest common hardware that fits is a RX 7900 XTX.

Can an RTX 4090 (24GB) run Qwen3 Coder 30B-A3B?

Yes. An RTX 4090's 24 GB runs Qwen3 Coder 30B-A3B at Q4_K_M (about 22 GB at 8K context).

Can a Mac run Qwen3 Coder 30B-A3B?

Yes — Apple Silicon with 32 GB of unified memory or more (macOS lets the GPU use ~75% of it, ~24 GB) runs Qwen3 Coder 30B-A3B at Q4_K_M.

How much VRAM does Qwen3 Coder 30B-A3B need at its full 256K context?

The KV cache grows with context: about 0.8 GB at 8K tokens versus 25.8 GB at the full 256K window (FP16 cache). Add that to the weights (19 GB at Q4_K_M) plus overhead — long contexts can cost more than a whole quantisation step.

Related

VRAM calculator for any model ·Token counter

Similar-size local models: Gemma 4 31B (31B) · Qwen3.6 27B (27B) · Qwen3.6 35B-A3B (35B) · Gemma 4 26B-A4B (26B) · GPT-OSS 20B (21B)

Last updated 2026-09-03. Architecture figures from the model's published config.json; see the methodology.