Featured:MeloLab– Create songs and background music with AI.

Qwen4 Max

Qwen4 Max on Together AI: list and promo pricing, 1M-token context, streaming-only requests, thinking defaults, image input, tool calling and caching limits.

Type:
AI Tool
Qwen4 Max product preview
Contents

What is Qwen4 Max?

Qwen4 Max on Together AI: list and promo pricing, 1M-token context, streaming-only requests, thinking defaults, image input, tool calling and caching limits.

Updated ·

Qwen4 Max on Together AI: Pricing, 1M Context, Streaming and Thinking

Announcement date:

Qwen4 Max is Alibaba's flagship Qwen model, and on Together AI it runs on Serverless Inference with a 1M-token context window. The practical points to know before you build: requests must stream, thinking is on by default and billed as output, prompt caching is automatic but best effort, and the model is serverless only with no fine-tuning. List pricing is $1.50 per 1M input tokens, $4.50 per 1M output tokens and $0.075 per 1M cached input tokens, with a 30% launch discount that ran through November 22, 2026.

Social announcement artwork reading Qwen4 Max is live on Together AI
Publisher announcement artwork; not a product interface screenshot.

What Qwen4 Max is

Qwen4 Max is the flagship model in Alibaba's Qwen family, released on November 16, 2026 and described by Alibaba as the most capable Qwen model to date. It is a sparse mixture-of-experts model with 2.4 trillion total parameters, of which 71 billion are active for each token.

On Together AI it is offered through Together Serverless Inference with a 1M-token context window, which makes it a candidate for long-document analysis, large codebases and multi-turn agent loops where context length is the binding constraint.

  • Flagship Qwen model, released November 16, 2026
  • Sparse mixture-of-experts: 2.4T total parameters, 71B active per token
  • 1M-token context window on Together Serverless Inference
  • Text and image input, text output

Pricing and the 30% launch discount

Together AI lists Qwen4 Max at $1.50 per 1M input tokens, $4.50 per 1M output tokens and $0.075 per 1M cached input tokens. During the launch promotion, which ran through November 22, 2026, those rates were cut by 30% to $1.05, $3.15 and $0.0525 respectively.

Because thinking tokens count as output tokens, the output rate is the number that most affects cost on reasoning-heavy workloads. If you need predictable spend, either cap output length or turn thinking off for tasks that do not need it. Whether the discount has been extended past November 22, 2026 is not stated in the source, so check current pricing before committing to a budget.

  • List: $1.50 input / $4.50 output / $0.075 cached input per 1M tokens
  • Promo through Nov 22, 2026: $1.05 / $3.15 / $0.0525 per 1M tokens
  • Thinking tokens are billed at the output rate
  • Discount extension beyond the stated date is unknown

Streaming is required, not optional

Qwen4 Max on Together AI accepts streaming requests only. If you send a request without "stream": true, the API returns a 400 error with the code streaming_required. This is a hard requirement rather than a performance preference, so any client, SDK wrapper or gateway in front of the model must be configured to stream.

In practice that means collecting the response from streamed chunks and assembling the final text yourself, and making sure intermediate layers such as proxies, serverless functions and logging do not buffer or drop the stream.

  • Always set "stream": true in the request body
  • A non-streaming call fails with 400 and code streaming_required
  • Check that proxies and function runtimes forward chunks rather than buffering
  • Assemble the final answer from streamed deltas on the client side

Thinking mode: on by default, billed as output

Thinking is enabled by default for Qwen4 Max, and thinking tokens are counted as output tokens. For reasoning-heavy work that default is usually what you want; for classification, extraction or short rewrites it adds latency and cost without much benefit.

To turn thinking off, pass reasoning={"enabled": False} in the Python SDK, or "reasoning": {"enabled": false} in the raw request body. Treat this as a per-request setting and apply it selectively rather than globally, so you keep reasoning where it earns its tokens.

  • Default: thinking on, tokens billed as output
  • Python SDK: reasoning={"enabled": False}
  • Raw request body: "reasoning": {"enabled": false}
  • Consider disabling for short, deterministic tasks

Image input, tool calling and prompt caching

Qwen4 Max accepts text and image input on Together AI and returns text. Each image is sent as an image_url block inside the message content, following the familiar multimodal message shape.

Tool calling works through the standard tools parameter, and the model can return more than one call in a single turn, so your tool-execution loop should handle parallel calls rather than assuming one call per response. Prompt caching is automatic, but it is best effort: repeated prefixes are often cached, not guaranteed to be, so do not build a cost model that assumes every repeat is a cache hit.

  • Multimodal input: text plus images as image_url blocks
  • Tool calling via the standard tools parameter
  • Multiple tool calls can arrive in one turn
  • Prompt caching is automatic but best effort

Availability limits and how to start

Qwen4 Max is serverless only on Together AI today and is not available for fine-tuning, so there is no dedicated-endpoint or custom-weights path for this model right now. Plan around the serverless API rather than a provisioned deployment.

A sensible first step is a small streaming request with thinking left on, to confirm your client handles streamed output and reasoning fields correctly. Then test a tool-calling turn with two tools available, and a request with an image block, before moving real traffic over. Alibaba has said smaller open-weight models from the Qwen4 series will follow, which may eventually change the deployment options available.

  • Serverless only; no fine-tuning for Qwen4 Max
  • No dedicated endpoint option documented for this model
  • Start with a streaming smoke test, then tool calling, then image input
  • Smaller open-weight Qwen4 models are expected to follow

Questions

How much does Qwen4 Max cost on Together AI, and is the 30% discount still active?

List pricing is $1.50 per 1M input tokens, $4.50 per 1M output tokens and $0.075 per 1M cached input tokens. A 30% launch discount ran through November 22, 2026, bringing those rates to $1.05, $3.15 and $0.0525. The source does not say whether the discount was extended, so confirm current pricing before budgeting.

Why does a Qwen4 Max request on Together AI return a 400 streaming_required error?

Qwen4 Max on Together AI accepts streaming requests only. A request sent without "stream": true is rejected with a 400 error carrying the code streaming_required. Set streaming on in the request body and make sure any proxy or wrapper passes chunks through.

How do I turn off thinking for Qwen4 Max, and are thinking tokens billed as output?

Yes, thinking tokens count as output tokens. Thinking is on by default; disable it per request with reasoning={"enabled": False} in the Python SDK or "reasoning": {"enabled": false} in the request body.

Can I fine-tune Qwen4 Max or deploy it as a dedicated endpoint on Together AI?

No. Qwen4 Max is serverless only on Together AI today and is not available for fine-tuning, so there is no custom-weights or dedicated deployment path documented for it. Alibaba has said smaller open-weight Qwen4 models will follow.

Original sources

  1. www.together.ai — Original source 1 ↗

Tool Info

Has this tool changed? Help us keep it up to date

See also: related tools and guides

Loading reviews…

Qwen4 Max Reviews

Not rated yet0 user reviews

Do you recommend Qwen4 Max? Leave a comment below!

Your rating
0 / 2000

Loading reviews…

Categories

FOR PRODUCT BUILDERS

Built something useful? Help people find it.

Start with your website. Add your logo, screenshots and product details to a listing you can keep up to date.

Verify your website badge for a free listing.