LANGUAGE MODEL Alibaba Last updated:

Qwen3-Coder

Open-Source Coding Agent Frontier

Qwen3-Coder is Alibaba's specialized coding-agent variant of Qwen 3, fine-tuned on real software-engineering traces. It edits code across repositories up to 1 million tokens long and can drive a terminal, not just complete a single function. Released open-weights under Apache 2.0 — the strongest open coding model of mid-2025.

Intelligence
Medium
Speed
Good
147 tok/s output
Cost
Low
$0.11 in / $0.80 out
Context
1M
Up to 1,000,000 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on Artificial Analysis. Cost buckets are fixed dollar thresholds keyed off output-token price ($/M out).
Intelligence
  • Top 1%≤ 1%
  • Top 5%≤ 5%
  • Top 10%≤ 10%
  • Good≤ 25%
  • Medium≤ 50%
  • Below avg> 50%
Speed
  • Top 1%≥ 345 tok/s
  • Top 5%≥ 237 tok/s
  • Top 10%≥ 196 tok/s
  • Good≥ 146 tok/s
  • Medium≥ 90 tok/s
  • Slow< 90 tok/s
Cost
  • Freeopen weights · self-host
  • Low< $1 / M out
  • Moderate$1–5 / M out
  • High≥ $5 / M out

Why it matters

Made "agentic coding under Apache 2.0" a real product, not a benchmark game. Cursor, Cline, Roo, and Aider all shipped Qwen3- Coder integrations within weeks — the first open model to drive serious agentic coding workflows.

Core Capabilities

Long Documents
Handles entire codebases, books, and multi-doc RAG.
Generative
Produces images, video, audio, or other media.
Coding
Strong real-world software engineering.
Agent Workflows
Built for tool use and autonomous tasks.

Context Window

1M tokens
≈ entire codebase
4k Chat 聊天
32k Long docs 长文档
128k Books 整本书
400k Multi-doc 多文档
1M This model 本模型
10M

Availability

API
Available
Product / App
Not available
Open Source
Released
Enterprise
Contact sales

Pricing Model

Free / self-host
Open weights — pay only for compute.
Self-host

Capability / Performance

Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).

Lower 20% Upper 80% This model
Reasoning
AA Intelligence Index · scaled to 10
1.7
5.6
4.0
Coding
SciCode · scaled to 10
1.8
4.3
3.2
Agentic tasks
Terminal-Bench Hard · scaled to 10
0.2
3.6
1.8
Context / memory
Context window size · log-scaled
6.0
9.0
9.0
Cost efficiency
Input price ($/M tokens) · cheaper scores higher
6.2
10.0
10.0
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.

What it feels like

  • Coding-specialist variant — large-context coder.
  • Best open-source reasoning model at its release — 235B-A22B (Thinking) beats DeepSeek-R1 on 17/23 benchmarks
  • Toggle-able thinking mode: same weights serve both reasoning and fast-chat modes
  • Strong 119-language coverage; the most genuinely multilingual frontier-tier model

Best use cases

  • Multilingual production deployments (119 languages) where most models stay English-centric
  • Self-hosted reasoning workflows that need both 'fast mode' and 'thinking mode' from one weight set
  • Open-weights agentic coding (Qwen3-Coder) with very large context windows

Tools to try

Not ideal for

  • Multimodal tasks (Qwen3 base is text — vision lives in Qwen3-VL, audio in Qwen3-Audio)
  • Edge / single-consumer-GPU deployments at the 235B scale

Model Evolution

View full evolution tree →