Qwen3 Coder Next GPU requirements: 39.8 GB at 4-bit, 159 GB at BF16
Qwen3 Coder Next needs 159 GB of GPU memory for its weights at BF16, 79.7 GB at FP8 and 39.8 GB at 4-bit, plus 0.81 GB of KV cache for a 32K context. The cheapest node that runs it here is 1× A40 at 4-bit, for $0.06 an hour on spot.
Weights plus a 32K 16-bit KV cache, with 10% of the node's memory kept free. Billed per minute.
- Parameters
- 80B3B active per token
- Weights, 4-bit
- 39.8 GBBF16: 159 GB
- KV cache, 32K
- 0.81 GBOne sequence, 16-bit cache
- Context window
- 256KNative
How much VRAM Qwen3 Coder Next needs, format by format.
Weights are parameters × bytes per parameter. Each context column adds the KV cache of one sequence of that length.
| Format | Weights | + 8Kcontext | + 32Kcontext | + 128Kcontext | + 256Kcontext | Cheapest node, 32K |
|---|---|---|---|---|---|---|
| BF16Published weights, fine-tuning | 159 GB | 160 GB | 160 GB | 163 GB | 166 GB | 4× A40 · $0.24/h |
| FP8Hopper, Ada, Blackwell, MI300 | 79.7 GB | 79.9 GB | 80.5 GB | 82.9 GB | 86.1 GB | 2× L40 · $0.18/h |
| 4-bitAWQ, GPTQ, GGUF Q4 | 39.8 GB | 40.0 GB | 40.6 GB | 43.1 GB | 46.3 GB | 1× A40 · $0.06/h |
The serving engine adds its own buffers, and an 8-bit KV cache halves the context columns. We keep 10% of a node's memory free when we say a model fits. Speed is a separate question: it follows memory bandwidth and, for a mixture of experts, the active parameters.
GPUs that run Qwen3 Coder Next, cheapest first.
For each format, the smallest node of each GPU model that holds the weights and a 32K context, sorted by spot price. One row per GPU model; larger nodes of the same GPU work too.
| Format | GPU | GPU memory | Spot | On-demand | 730 h spot | Action |
|---|---|---|---|---|---|---|
| 4-bit | 1× A40Ampere · 40.6 GB needed | 48 GB | $0.06 | $0.09 | $43.80 | Deploy |
| 4-bit | 1× L40Ada Lovelace · 40.6 GB needed | 48 GB | $0.09 | $0.19 | $65.70 | Deploy |
| 4-bit | 1× RTX 6000 AdaAda Lovelace · 40.6 GB needed | 48 GB | $0.09 | $0.19 | $65.70 | Deploy |
| 4-bit | 1× RTX A6000Ampere · 40.6 GB needed | 48 GB | $0.09 | $0.17 | $65.70 | Deploy |
| FP8 | 2× L40Ada Lovelace · 80.5 GB needed | 96 GB | $0.18 | $0.38 | $131.40 | Deploy |
| FP8 | 2× RTX 6000 AdaAda Lovelace · 80.5 GB needed | 96 GB | $0.18 | $0.38 | $131.40 | Deploy |
| FP8 | 1× RTX PRO 6000 BlackwellBlackwell · 80.5 GB needed | 96 GB | $0.19 | $0.39 | $138.70 | Deploy |
| FP8 | 4× L4Ada Lovelace · 80.5 GB needed | 96 GB | $0.20 | $0.32 | $146.00 | Deploy |
| BF16 | 4× A40Ampere · 160 GB needed | 192 GB | $0.24 | $0.36 | $175.20 | Deploy |
| BF16 | 4× L40Ada Lovelace · 160 GB needed | 192 GB | $0.36 | $0.76 | $262.80 | Deploy |
| BF16 | 4× RTX 6000 AdaAda Lovelace · 160 GB needed | 192 GB | $0.36 | $0.76 | $262.80 | Deploy |
| BF16 | 4× RTX A6000Ampere · 160 GB needed | 192 GB | $0.36 | $0.68 | $262.80 | Deploy |
Spot capacity can be reclaimed with a 2-minute notice; for a live endpoint, use on-demand. How spot works · serving a language model.
What a long context costs in memory.
Hybrid: 12 full-attention layers (2 KV heads of 256), the other 36 are linear-attention (Gated DeltaNet) layers with a fixed-size state.
| Context | KV cache, 1 sequence | 8 sequences |
|---|---|---|
| 8K tokens | 0.2 GB | 1.61 GB |
| 32K tokens | 0.81 GB | 6.44 GB |
| 128K tokens | 3.22 GB | 25.8 GB |
| 256K tokens | 6.44 GB | 51.5 GB |
Serving several users
KV cacheEvery concurrent request keeps its own cache, so memory for context grows with users × context length. A server batching 8 requests of 32K needs eight times the single-sequence figure — 6.44 GB here.
- Recommended node1× A40
- Weights at 4-bit39.8 GB
- Context that fits, 1 sequence134K
Paged attention (vLLM, SGLang) allocates the cache in blocks as requests grow, so a node holds more short requests than the worst case suggests.
Qwen3 Coder Next, as published.
- Developer
- Alibaba QwenCoding model on the Qwen3-Next hybrid architecture
- Parameters
- 80B total, 3B active per token
- Architecture
- Mixture of experts, 48 layersHybrid: 12 full-attention layers (2 KV heads of 256), the other 36 are linear-attention (Gated DeltaNet) layers with a fixed-size state
- Context window
- 262,144 tokensAs set in the published configuration
- Published format
- BF16, 159 GB
- Licence
- Apache 2.0
- Released
- 30 January 2026
- Source
- Qwen/Qwen3-Coder-Nextconfig.json read 26 October 2026
Serve it with vLLM
template: vLLMOn 1× A40, from the instance terminal:
$ vllm serve Qwen/Qwen3-Coder-Next \ --tensor-parallel-size 1 \ --max-model-len 32768
Very recent architectures need a recent vLLM; the model card gives the minimum version and any extra flags.
Qwen3 Coder Next on a GPU — the questions that come up
How much VRAM does Qwen3 Coder Next need?
For the weights alone: 159 GB at BF16, 79.7 GB at FP8 and 39.8 GB at 4-bit. Add 0.81 GB of KV cache for one 32K-token context, or 6.44 GB at the full 256K window, and keep about 10% of the memory free for activations.
Can Qwen3 Coder Next run on a single GPU?
One RTX 4090 (24 GB): no. One RTX 5090 (32 GB): no. One H100 SXM (80 GB): yes, at 4-bit. One H200 SXM (141 GB): yes, at FP8. One B200 SXM (180 GB): yes, at BF16. Those answers assume a 32K context and the widest format that still fits.
What is the cheapest way to run Qwen3 Coder Next?
Here, 1× A40 at 4-bit: $0.06 per hour on spot and $0.09 on-demand, billed per minute — $43.80 for a 730-hour month on spot. That node has 48 GB of GPU memory for the 40.6 GB this setup needs.
Does Qwen3 Coder Next need less memory because only 3B parameters are active?
No. A mixture-of-experts model routes each token through a few experts, but every expert has to be resident in GPU memory, so the memory bill follows the 80B total. What the 3B active parameters change is the work per token: generation reads far less of the weights each step than a dense model of the same size would, which is why it is fast for its size.
Which format should I run it in?
BF16 keeps the published weights untouched and is the format for fine-tuning. FP8 halves the memory and runs in hardware on Hopper, Ada, Blackwell and MI300-class GPUs. 4-bit quantisation (AWQ, GPTQ, GGUF) quarters it, at a quality cost that is larger on small models than on big ones; check it against your own evaluation before relying on it.
How do I deploy it?
Pick the node in the table above and launch it with the vLLM template; the model downloads from Hugging Face (Qwen/Qwen3-Coder-Next) on first start. Spread the weights over the node with --tensor-parallel-size equal to the GPU count, and set --max-model-len to the context you need. New architectures need a recent vLLM: check the model card for the minimum version.
Sizing another model? GPU requirements for 64 open models · serving a language model · fine-tuning.
Qwen3 Coder Next on 1× A40 for $0.06 an hour.
Launch the node with the vLLM template, point it at Qwen/Qwen3-Coder-Next, stop it when you are done. Prepaid, billed by the minute.
Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription.