Qwen3-Coder-Next (80B-A3B)
Qwen3-Coder-Next (80B-A3B) by Alibaba Qwen (80B parameters, MoE with 3B active per token). At Q4_K_M with an 8K context it needs about 48.1 GiB of memory. An 80B coding MoE with only 3B active, and just 12 of its 48 layers keep a KV cache, so long contexts stay cheap. At 4-bit a 64 GB machine is borderline; 96 GB is comfortable.
Last updated: 2026-10-13
Model facts
| Developer | Alibaba Qwen |
|---|---|
| Parameters | 80B (3B active, MoE)[src] |
| Native context | 262,144 tokens[src] |
| Licence | Apache-2.0[src] |
| Commercial use | Yes |
| KV cache | 12 of 48 layers (the rest are linear attention with a fixed-size state)[src] |
| Released | 2026-02 |
Good for: coding · agentic coding · long context · 64GB+ unified-memory machines
Run it
ollama run qwen3-coder-next:q4_K_M
Memory needed
Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.
| Quantization | 4K context | 8K context | 32K context |
|---|---|---|---|
| Q4_K_M | 48.0 GiB | 48.1 GiB | 48.7 GiB |
| Q5_K_M | 56.2 GiB | 56.3 GiB | 56.9 GiB |
| Q6_K | 65.0 GiB | 65.1 GiB | 65.7 GiB |
| Q8_0 | 83.7 GiB | 83.8 GiB | 84.4 GiB |
| FP16 | 157.1 GiB | 157.1 GiB | 157.7 GiB |
Which hardware runs it
Best quantization that fits each device with a 8K context. Estimates, not benchmarks. See the methodology.
Frequently asked questions
How much VRAM does Qwen3-Coder-Next (80B-A3B) need?
With an 8K context, about 48.1 GiB at Q4_K_M and 83.8 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.
Can Qwen3-Coder-Next (80B-A3B) run on an GeForce RTX 4090?
Qwen3-Coder-Next (80B-A3B) does not fit on the GeForce RTX 4090 (24 GB), even at Q4_K_M with an 8K context.
Can I run Qwen3-Coder-Next (80B-A3B) on a Mac?
Yes. Qwen3-Coder-Next (80B-A3B) fits on the M3 Ultra (96 GB) at Q6_K with an 8K context, using about 65.1 GiB, at an estimated 215 tokens/s.
Can Qwen3-Coder-Next (80B-A3B) be used commercially?
Qwen3-Coder-Next (80B-A3B) is released under Apache-2.0. The licence allows commercial use (read its conditions).
Sources
- https://docs.aws.amazon.com/bedrock/latest/userguide/model-card-qwen-qwen3-coder-next.md
- https://huggingface.co/Qwen/Qwen3-Coder-Next/blob/main/LICENSE
- https://huggingface.co/Qwen/Qwen3-Coder-Next
- https://huggingface.co/Qwen/Qwen3-Coder-Next/resolve/main/config.json
- https://huggingface.co/api/models/Qwen/Qwen3-Coder-Next?expand[]=safetensors
- https://ollama.com/library/qwen3-coder-next/tags