Skip to content

AI Tools

What Is Qwen4-Max? Features, Pricing, Release Date & Availability

Quick Answer

Qwen4-Max is Alibaba’s flagship AI model and the first of the Qwen4 family. It was a Token-Plan-only preview from 11 November 2026, and became generally available on 16 November 2026 as qwen4-max on Alibaba Cloud / QwenCloud with pay-as-you-go pricing of $1.50 / $4.50 per 1M input / output tokens. It is a sparse-MoE multimodal reasoning model with a 2.4-trillion-parameter backbone (71B active per token), a 1M-token context, and Alibaba-reported benchmark gains over Qwen3.8-Max — Terminal-Bench 4.0 up from 37.9 to 48.5, and average cost per task down 62% on Qwen’s own Work360 test. It is hosted-only: no weights or open-source licence for Max itself; smaller Qwen4 models are promised as open-weight soon. The old preview ID (qwen4-max-preview) still retires on 25 November 2026.

What Is Qwen4-Max?

Qwen4-Max is the top tier of Alibaba’s next-generation Qwen4 family. Alibaba’s official Qwen account announced it on X on 11 November 2026 at 02:01 UTC with the headline “Qwen4 is coming.” Only Qwen4-Max is out so far. It launched as a preview on 11 November 2026 and became generally available on 16 November 2026. The rest of the family — Qwen4-Plus, Qwen4-Flash and Qwen4-27B, named at Alibaba’s Apsara Conference on 29 October — has no release date yet.

Key Features

1. 1-million-token context window

Published in Alibaba’s documentation (1,000,000 tokens), with a maximum output of 393,216 tokens.

2. Multimodal input

Text, image and video input, with thinking on by default.

3. Built for coding and agents

Alibaba pitches it at coding, agentic work, front-end builds and professional software workflows, and says it beats Qwen3.8 on world knowledge, code generation and sustained performance on long tasks with repeated tool calls. Treat that as Alibaba’s own claim: Alibaba has since published its own benchmark figures (see the comparison below), but no model card and no independent benchmark yet.

4. Works inside coding tools

Through the Token Plan it can be used inside Claude Code, Qwen Code, Cursor, Codex, OpenCode, Cline and similar tools via OpenAI-compatible and Anthropic-compatible endpoints.

5. Model ID

qwen4-max is the generally available model ID. qwen4-max-preview is the preview ID, which retires on 25 November 2026.

Pricing

Qwen4-Max now has pay-as-you-go pricing on Alibaba Cloud Model Studio / QwenCloud: $1.50 per 1M input tokens and $4.50 per 1M output tokens (international, Singapore rates), with implicit cache reads at $0.075 per 1M tokens. Domestic rates are ¥10 / ¥30 per 1M input / output tokens, with ¥0.5 cache.

Pay-as-you-go Input (per 1M tokens) Output (per 1M tokens) Implicit cache (per 1M tokens)
International $1.50 $4.50 $0.075
Domestic ¥10 ¥30 ¥0.5

The Token Plan below still applies to the preview / coding-tool route.

Token Plan Monthly price Credits per month First-time price
Lite $6 20,000 $3 until 30 Nov 2026
Essential $10 45,000 $5 until 30 Nov 2026
Standard $18 80,000 —
Pro $43 200,000 —

During the preview, Qwen4-Max uses credits at 20% of the standard rate (a limited-time 80% discount, as shown on the Token Plan page). Team plans run $20–$180 per seat per month. There is no free trial on the Token Plan.

What Changed on 16 November 2026

  • Preview → generally available: qwen4-max on Model Studio, pay-as-you-go; the preview ID retires on 25 November 2026.
  • First real specs: 2.4T backbone / 71B active / 328B N-gram embeddings.
  • First benchmarks (Alibaba’s own): Terminal-Bench 4.0 48.5, Work360 52.69, CodeArena WebDev No. 6.
  • First per-token price, and a 62% cost-per-task cut vs Qwen3.8-Max on Work360.
  • Confirmed closed: Max is hosted-only; open-weight applies to smaller Qwen4 models, undated.

Release Date and Availability

  • Announced: 11 November 2026 (02:01 UTC / 10:01 Beijing).
  • Generally available: 16 November 2026 as qwen4-max on Alibaba Cloud Model Studio / QwenCloud, pay-as-you-go.
  • Preview window: 11 November to 25 November 2026 (16:00 UTC / 24:00 Beijing time). Alibaba says the preview ID will be retired after the preview ends; the generally available model qwen4-max continues.
  • Where to try it: Token Plan (qwencloud.com/pricing/token-plan), Qwen Studio at chat.qwen.ai (in the model picker, next to Qwen3.8-Max), QwenWork, Qoder and the Qwen app. The four apps are described as “early access”, so availability inside each one may roll out gradually.
  • Limits: Token Plan use is for interactive use in compatible AI programming and agent tools only — it cannot be used for automated scripts or application backends, and personal use is allowed only with interactive tools.

Qwen4-Max vs Qwen3.8-Max

Now a real comparison is possible: same 2.4T backbone size as Qwen3.8-Max, but 25% fewer active parameters (71B), ~50% fewer tokens per Terminal-Bench attempt (289K → 151K), score 37.9 → 48.5, and Work360 cost per task $4.43 → $1.67. All figures are Alibaba’s own.

How to Use Qwen4-Max

  1. Easiest way — Qwen Studio: go to chat.qwen.ai, sign in, and pick Qwen4-Max from the model picker.
  2. For coding tools — Token Plan: subscribe at qwencloud.com/pricing/token-plan (from $3 for first-time Lite buyers until 30 November), then point your coding tool (Claude Code, Cursor, Qwen Code and others) at the Token Plan endpoint.
  3. In Alibaba’s apps: open QwenWork, Qoder or the Qwen app and look for Qwen4-Max in early access.

What Is Still Unknown

  • No independent benchmarks yet (Artificial Analysis / LMArena had no score at last check).
  • No model card, weights or licence for Max; no date/sizes for the promised open-weight smaller models.
  • No word on what happens to Token Plan pricing for the GA model vs the preview.

Frequently Asked Questions

What is Qwen4-Max?

Alibaba’s new flagship model and the first of the Qwen4 family, launched as a preview on 11 November 2026 and generally available from 16 November 2026, for coding and agentic work.

When was Qwen4-Max released?

11 November 2026 as a preview, then generally available on 16 November 2026 as qwen4-max. The rest of the Qwen4 family has no release date yet.

How much does Qwen4-Max cost?

$1.50 / $4.50 per 1M tokens pay-as-you-go; Token Plan $6–$43/month remains an option.

Is Qwen4-Max free?

No. The Token Plan has no free trial. Qwen Studio has a free plan, but features and allowances differ by service.

What is the context window?

1,000,000 tokens, with up to 393,216 output tokens.

What are Qwen4-Max’s specs?

2.4T backbone, 71B active per token, 1M-token context.

How is it different from Qwen3.8-Max?

Alibaba reports it is stronger on knowledge, code and long agentic tasks, with a 1M-token context, Terminal-Bench 4.0 up from 37.9 to 48.5 and ~50% fewer tokens per attempt. All figures are Alibaba’s own; no independent benchmarks exist yet.

Is Qwen4-Max open source?

No. Max is hosted-only; smaller Qwen4 models are promised as open-weight soon, with no date.

When does the Qwen4-Max preview end?

25 November 2026 — but only the preview ID retires; the GA model qwen4-max continues.

Final Verdict

The preview’s open questions are now answered on specs, price and benchmarks — and the answers are strong on efficiency (fewer active parameters, half the tokens, 62% cheaper per task on Alibaba’s own test). The caveats moved, not vanished: every benchmark is Alibaba’s own, Max itself is closed, and the open-weight promise has no date. Worth trying on a real coding/agent task via the pay-as-you-go API; don’t buy it as an open model.

Editorial update: November 16, 2026. Updated for the general-availability launch — specs, pricing and benchmarks added; preview details kept where they still apply. Benchmarks are Alibaba’s own and not yet independently verified.