Alibaba's hosted coding model — a million tokens of context aimed at agents that edit whole repositories rather than single files.

Key specifications

Capability
47.4
Context window
1.0M
Max output
65.5K
Input $/1M
$1.00
Output $/1M
$5.00
License
Proprietary

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Text input

Produces

  • Text output

Capabilities

Function callingOpen weights

What is Qwen3 Coder Plus?

There are two things called Qwen3-Coder, and picking the wrong one costs either money or a weekend.

Qwen3 Coder Plus is Alibaba's hosted, closed-weight coding model, reached through an API. Qwen3-Coder is the open-weight family you can download — 480B-A35B, 30B-A3B and the Next variants, all published to Hugging Face and ModelScope. Same lineage, different products. If you came here planning to self-host, the model you want is the second one.

What a million tokens changes

Plus exposes a 1,048,576-token context window. For coding work that is a category shift rather than a bigger number.

At 8K you paste a function. At 128K you paste a few files and hope you picked the right ones. At a million you hand over the repository — source, tests, migrations, config — and ask a question that depends on how those pieces actually connect. The open Qwen3-Coder models ship with 256K natively and stretch toward 1M using YaRN extension; the hosted Plus endpoint gives you the long window without the deployment work.

That matters most for the failures that are not local. A test that breaks only in CI. A type that drifted three refactors ago. Nothing in the file you are looking at explains either.

Pricing swings more than the model does

This is the part worth reading twice. Across providers, Plus runs anywhere from $0.57 to $6.00 per million input tokens, and $2.29 to $60.00 per million output tokens.

That is roughly a tenfold spread on input and twenty-sixfold on output — for the same model. Where you route matters more than most people's model choice does. Before committing a coding agent to production traffic, price your actual token mix against your actual provider, because a codebase-sized prompt multiplies that difference by every call your agent makes.

It answers, it does not deliberate

Plus does not support a reasoning mode. It responds directly rather than working through a visible chain of steps.

For most coding work that is fine, and it is faster. Where it shows is architectural judgment — untangling a race condition, choosing between two migration strategies. There, a model that reasons before answering earns its latency, and pairing Plus for bulk edits with a reasoning model for the hard calls tends to beat using either alone.

Specifications

Field Value
Provider Alibaba
Context window 1,048,576 tokens
Maximum output 65,536 tokens
Input text
Output text
Tool calling Supported
Reasoning mode Not supported
Weights Closed (hosted API)
Knowledge cutoff April 2025
Released 23 July 2025

Provider pricing moves, and the spread above is wide enough to matter. Confirm your own rate before you budget.

See how it compares in the AI model directory, or look at Gemini 2.5 Pro if you need reasoning over the same context size.

How to evaluate Qwen3 Coder Plus

Use the recorded facts as a shortlist, then validate the model against representative inputs and production constraints.

Workload fit

Qwen3 Coder Plus is categorized for Text generation, Text-to-text AI models. Its current record accepts text and produces text. Confirm file formats, preprocessing, and provider-specific request schemas before implementation.

Capacity and cost

The directory records 1.0M context and 65.5K maximum output. Listed token prices are $1.00 input and $5.00 output per 1M tokens. Treat missing values as unknown and recheck current commercial terms.

Operational behavior

Recorded output speed is — and time to first token is —. Hosting route, region, prompt length, concurrency, and provider load can materially change both measurements.

Evidence boundary

This page separates sourced model facts from incomplete fields. Benchmark evidence is displayed only when its source is linked and verified. Before choosing Qwen3 Coder Plus, test task quality, tool reliability, safety behavior, data controls, rate limits, and total cost on the exact route you intend to use.

Strengths

  • + 1,048,576-token context holds a real repository, not just the file you are editing
  • + Built for agentic coding: tool calling is supported and the format targets coding agents
  • + Knowledge cutoff of April 2025 is recent for a coding model
  • + Open-weight Qwen3-Coder siblings exist if you need to self-host a similar capability

Limitations

  • No reasoning mode — it does not deliberate before answering
  • Closed weights: Plus is API-only, unlike the open Qwen3-Coder releases
  • Provider pricing varies roughly 10x on input and 26x on output — routing matters enormously
  • Text only, so no screenshot or diagram input when debugging UI work

Frequently asked questions

Qwen3 Coder Plus is a Alibaba model listed in the provider’s public registry. Hosted Qwen coder for software agents, repo edits, and long-context code

Related approved models in Text-to-text AI models. Compare specifications and verify fit for your workload.

Anthropic

Claude 3.5 Sonnet

The 2024 Sonnet that set the standard for practical coding work — now well behind the current Claude line at identical pricing.

200K contextCompare

OpenAI

GPT-4o

OpenAI's omni-era workhorse — text and image in, text out, with a cached-input rate that halves the cost of repeated context.

128K contextCompare

Meta

Llama 3.1 405B

Meta's largest open-weights model, competitive with frontier closed models.

128K contextCompare

DeepSeek

DeepSeek V4 Flash 0731

Official DeepSeek V4 Flash release with enhanced agentic capabilities and integrated DSpark speculative decoding

1M contextCompare