Back to the index
Alibaba Qwen · Qwen
No. 31

Qwen3-Coder-480B-A35B

Best open-weights Qwen for agentic coding — great value.

OpenOpen weightsText
Field notes
Context window
1.0Mtokens
Max output
66Ktokens
Input price
$0.22/ 1M tokens
Output price
$1.80/ 1M tokens
Cached input
$0.15/ 1M tokens
Modalities
Text
Knowledge cutoff
Apr 2025
Released
Jul 2025
Parameters
480B
License
Apache-2.0
Approx. speed
Overview

What this model is

Qwen3-Coder-480B-A35B-Instruct is Alibaba's flagship open-weights coding model, a 480B-parameter MoE with 35B active params and native long context (extendable toward 1M). It is the strongest Qwen for coding and agentic tool use, roughly comparable to Claude Sonnet 4 among open models, and ships under Apache-2.0. Long multi-file agent runs still need more supervision and can drift compared with the Claude/GPT frontier.

Strengths

  • Top-tier open-weights coding ability
  • Apache-2.0 license, self-hostable
  • Strong agentic tool-use for an open model
  • Native long context for whole-repo work

Trade-offs

  • Drifts on long multi-file agent sessions
  • Below frontier reliability without supervision
  • Text-only, no vision
  • Heavy to self-host (480B MoE)

Best for

  • Self-hosted coding assistants
  • Agentic code generation on a budget
  • Repo-scale code understanding and refactoring
  • Tool-use and browser automation

Not ideal for

  • Unsupervised long-horizon refactors
  • Vision or multimodal tasks
Capabilities

Capability profile

Normalized 0–100 scores, comparable across the whole catalog.

Reasoning72Coding78Math74Writing70Knowledge76Speed62Agentic72Vision5Multilingual80Long Context82
Reasoning
72
Coding
78
Math
74
Writing
70
Knowledge
76
Speed
62
Agentic
72
Vision
5
Multilingual
80
Long Context
82
Benchmarks

How it scores

Public benchmark results, with independent third-party results where available. Bars normalize percentages to 100 and Elo ratings to a 1500 ceiling.

BenchmarkResult
SWE-bench Verifiedindependent
CodingSWE-bench
67%
SWE-bench Proindependent
CodingSWE-bench Pro
38.7%
Aider Polyglotindependent
CodingAider
61.8%
Alternatives

Comparable models

Models in a roughly similar class — worth weighing against this one.