Qwen3-Coder-Next

Overview. Qwen3-Coder-Next is an open-weight, code-specialised model from Qwen, Alibaba's model team, released in February 2026 under the Apache-2.0 licence. It is a Mixture-of-Experts design with 80B total parameters but only 3B active per token, drawn from 512 experts, and it carries a native 262,144-token context window. Developers reach for it when they want a permissively licensed coding model that handles whole repositories in context without the compute footprint of a dense 80B model.

Key strengths. The model is built for coding agents and local development, and the benchmark signal backs that focus: its Artificial Analysis Coding Index of 36.2 has held steady across repeated measurements, a solid result for a model activating just 3B parameters. It ships with documented tool-calling support, so it can drive structured function calls inside an agent loop. Median throughput of roughly 110 tokens per second keeps Inference fast enough for IDE-style interaction, and the 262K context lets it take in large codebases in a single request.

When to use it. Use this for code generation, refactoring and repository-scale code understanding when you need a large context window and an open licence rather than top-end general reasoning. It fits interactive workloads such as IDE assistants and coding chat, where its throughput matters more than a sub-second first token. Its benchmark profile, with a Coding Index well above its general Intelligence Index, makes it a sensible pick for programming tasks that don't require extended deliberation.

Limitations. The model runs in non-thinking mode only, with no extended reasoning or chain-of-thought blocks, and its general-purpose Intelligence Index (10.1) and Agentic Index (3.6) on Artificial Analysis are low, so sustained multi-step autonomous tool loops and non-code tasks are weak spots. It is text-only, with no image, audio or document input, and its training cutoff is not published. Median time to first token sits above one second, which rules it out for latency-critical realtime use.

Sovereignty note. Qwen3-Coder-Next is built by Qwen, Alibaba's model team in China, and its weights are open under Apache-2.0, so it can be run on infrastructure of your choice. Where requests run, and whether inputs are retained or used for training, depends on the endpoint serving it; see the endpoint facts on this page.