Can I run Qwen3 Coder Next 80B-A3B?

Qwen3 Coder Next 80B-A3B is a coding specialist built as a Mixture-of-Experts, and that architecture is the whole story. It has 80B total parameters but only activates 3B per token, so it runs much faster than its size suggests while still needing memory for the full model. At 4-bit it lands around 48.5 GB, which rules out every consumer GPU: it does not fit on a 24 GB RTX 4090, let alone a 12 GB 3060. The realistic home for it is an Apple Silicon Mac with plenty of unified memory, or a workstation with 64 GB or more of system RAM. Plan around the minimum 64 GB figure, not the active 3B.

In daily use the MoE design pays off. On an Apple M Max it streams at roughly 192 tokens per second, which feels instant for code completion and refactoring, and even pure CPU inference on DDR5 manages about 28 tokens per second, slow but genuinely usable for a model this large. The 256K context window is the headline feature for working across whole repositories, but memory is the catch: at 128K context the total footprint climbs to about 95.8 GB. Keep that in mind before you load a massive codebase, because the KV cache, not the weights, is what will push you over the edge on a 64 GB machine.

Against the dense alternatives in its weight class, like Llama 3.1 70B, this model generally trades raw breadth for coding focus and speed: the MoE routing means it answers faster than a 70B dense model while specialising in code rather than general chat, where the smaller Qwen 3 chat variants are a better fit. Its standout trait is that speed-to-size ratio, getting near-instant generation from an 80B-class model. And the practical bonus is the license: Apache 2.0, so you can use it freely in commercial and production work with no provider-specific restrictions to read through first.