From an 8GB laptop card to a 24GB desktop, here's exactly which Qwen3-Coder you can run — and why 24GB is the sweet spot for a genuinely good local coding assistant.
Yes — and the variant you can run comes down to [VRAM](/how-much-vram-do-you-need-to-run-ai-models-2026). Here's the quick map: the 8B fits on an 8GB card; the 30B-A3B (a MoE model) wants roughly 18GB of combined memory and is ideal on a [24GB GPU](/best-gpu-for-qwen3-coder-2026) where it fits entirely in VRAM; the 32B dense model targets 24GB at [Q4](/which-quantization-should-you-use-q4-q5-q8-2026) (a card like an RTX 4090); and the 480B-A35B flagship needs a multi-GPU server, not a home rig. So the honest headline: 24GB is the sweet spot for a genuinely good local coding assistant, 8GB gets you the lightweight 8B, and the giant flagship is out of home reach. Here's the detail for each.
The numbers for each variant
Let's be specific. The 30B-A3B is the one most people ask about, and it's a MoE with only ~3B active parameters per token — so it runs fast, but you still need to fit all 30B in memory. At 4-bit, it wants about 18GB of combined memory (VRAM plus system RAM) for a usable ~6+ tokens/second if you offload — but the good experience is fitting it entirely in VRAM on a [24GB card](/best-gpu-for-qwen3-coder-2026) (an RTX 3090 or 4090), where it's genuinely quick. The 32B dense model is the quality step up for complex, multi-file tasks; at [Q4](/which-quantization-should-you-use-q4-q5-q8-2026) it also targets 24GB and runs best on an RTX 4090-class card. The 8B is the accessible entry — it fits 8GB and is fine for autocomplete and small edits, if not the heavy repository reasoning. And remember Qwen3-Coder's headline 256K [context](/what-is-a-context-window-explained-2026): using a lot of it grows the KV cache and eats VRAM, which is another reason 24GB (not 16GB) is the comfortable target for serious coding. [Q4_K_M](/which-quantization-should-you-use-q4-q5-q8-2026) is the best quality-per-GB for these models.
24GB is the sweet spot: it fits the 30B-A3B or 32B in VRAM with room for Qwen3-Coder's big context. · Unsplash
So what should you run?
Match it to what you've got. If you have a [24GB GPU](/best-gpu-for-qwen3-coder-2026) — an RTX 4090 or a used 3090 — run the 30B-A3B (fast, capable) or the 32B (a touch smarter on hard multi-file work); this is the setup that gives you a genuinely good local coding assistant. If you have 8-16GB, run the 8B for autocomplete and lighter help, and keep your context modest. If you're buying, a used RTX 3090 is the cheapest route to 24GB and the best value for this exact job — see our best GPU for Qwen3-Coder pick. Don't chase the 480B flagship at home; it's a data-center model, and the 32B gets you most of the way for a fraction of the hardware. Before you commit, [size your exact variant and context in the VRAM calculator](/tools/vram-calculator) — it tells you instantly whether your card is comfortable. Then set it up in VS Code and you've got a private coding assistant running on your own machine.
Quick answers
How much VRAM do you need for Qwen3-Coder?
It depends on the variant. The 8B fits on an 8GB GPU. The 30B-A3B (a Mixture-of-Experts model) needs about 18GB of combined VRAM and system RAM at 4-bit for usable speed, but it's ideal fully in VRAM on a 24GB card. The 32B dense model targets 24GB at Q4 quantization. The 480B-A35B flagship requires a multi-GPU server. For a genuinely good local coding assistant, 24GB is the sweet spot — it fits the 30B-A3B or 32B in VRAM with room for context. Also remember that Qwen3-Coder's large 256K context window grows the KV cache and uses additional VRAM as you fill it.
Can Qwen3-Coder run on 8GB?
Yes, but only the 8B variant. On an 8GB GPU, Qwen3-Coder 8B runs and is useful for autocomplete and smaller coding tasks, though it lacks the heavy repository-level reasoning of the larger variants. The 30B-A3B and 32B models need roughly 24GB of VRAM to run well, so they're out of reach on 8GB. If you're serious about a local coding assistant and can upgrade, a 24GB card (a used RTX 3090 is the cheapest route) unlocks the far more capable 30B-A3B and 32B. On 8GB, keep your context modest to avoid running out of memory.
What is the best Qwen3-Coder setup for local coding?
For most people, the 30B-A3B or 32B variant on a 24GB GPU (a used RTX 3090 for value or an RTX 4090 for speed), quantized to Q4_K_M, which is the best balance of quality and VRAM. That configuration fits the model entirely in VRAM, runs quickly, and leaves headroom for a decent slice of the 256K context. Serve it with Ollama or LM Studio and connect an editor extension like Continue (for chat and autocomplete) or Cline (for autonomous multi-file tasks) in VS Code. If you only have 8GB, use the 8B variant instead. Size your exact setup in a VRAM calculator before committing.
