AI Model Comparison

Model profile & tracker

Qwen 4 Max: Price, Context and Specs (Updated)

Qwen4-Max (released 16 November 2026): $1.50/$4.50 per million tokens, 2.4T-parameter MoE, 1M context. Specs, benchmarks, access and vs Hydrogen and Neon.

Verified 2026-11-22

Updated 22 November 2026. This page tracks Qwen4-Max — Alibaba's flagship released on 16 November 2026 — and gets rewritten when Alibaba ships the promised open weights, a Western cloud listing, or a price change.

Qwen4-Max is the cheapest closed flagship on the market as of its release week: a 2.4-trillion-parameter sparse mixture-of-experts backbone, about 71B parameters active per token, native image input and a 1M-token context window, listed at $1.50 input / $4.50 output per million tokens. The output price is 60% of Gemini 4 Hydrogen's $7.50 and 11% of GPT-6.2 Astra's $40.00.

The specs, as published

Spec Value
Model ID qwen4-max
Released 16 November 2026
Backbone 2.4T parameters, sparse MoE
Active per token ~71B (plus 328B n-gram embedding parameters)
Context window 1,000,000 tokens
Max input 991,808 tokens, or 983,616 with thinking on
Output cap 131,072 tokens, or 262,144 with thinking
Input price $1.50 / 1M tokens
Output price $4.50 / 1M tokens
Cached input $0.075 / 1M tokens
Explicit cache $1.875 / 1M (5-minute), $3.00 / 1M (1-hour)
Open weights Not released

Thinking is on by default, and up to 262K tokens of reasoning count against the combined thinking-plus-answer budget. That default is the number to watch when you estimate cost: every request pays for reasoning before it pays for the answer.

The cost story Alibaba published

The headline claim is a 62% drop in average cost per Work360 task: $4.43 on Qwen3.8-Max to $1.67 on Qwen4-Max. Two mechanics drive it: token usage drops roughly 24% (Xhigh effort) and 35% (High) on Work360, and 48–49% on Terminal-Bench 4.0; and the API price itself is below Qwen3.8-Max's, with the largest cut in cached input.

Benchmark Qwen3.8-Max Qwen4-Max
Work360 47.74 52.69
DeepSWE v1.1 69.3 75.7
Terminal-Bench 4.0 38.9 48.5

These are Alibaba's own published numbers, run by the provider on its own benchmark suite. Treat the direction — better scores on less money — as well supported, and the exact figures as unverified until an independent lab reproduces them. For how provider-run scores mislead, see the benchmark guide.

How to access it

QwenCloud is the entry point: the API at the prices above, or the Token Plan subscription (from $6 a month, with a promotional $3 tier for new users until 30 November 2026). The preview checkpoint qwen4-max-preview is Token-Plan only and uses credits at 20% of the standard rate during the preview. The model is also surfaced in QwenWork, Qoder and QwenStudio.

Qwen4-Max vs the other cheap flagships

vs Gemini 4 Hydrogen ($1.50 / $7.50): same input price, Hydrogen is 1.05M context to Qwen's 1M, but Qwen's output is 40% cheaper — $4.50 against $7.50. If you write more than you read, Qwen wins on bill; if your input is image, video, audio or PDF, Hydrogen takes all five modalities while Qwen4-Max takes text and image. Details on the Hydrogen page.

vs Gemini 4 Neon ($3.00 / $15.00): Qwen4-Max is half the input price and less than a third of the output price for a comparable window. Neon remains the default if your stack is already on Google.

vs the open-weight Qwen3.8 checkpoints: this is the trade-off. Qwen3.8-Max weights are downloadable; Qwen4-Max weights are not. If self-hosting or fine-tuning matters, the older model is still the one you can own.

FAQ

Is Qwen4-Max open source? No. The checkpoint is closed. Alibaba's open-weight Qwen line continues separately; when a Qwen4-Max-class open checkpoint ships, this page gets updated.

Is the $1.50 / $4.50 price confirmed? Yes, on QwenCloud's own listing, read 22 November 2026. Early coverage attributed the figures to press reports before the listing was live; the listing now confirms them.

What is the context window? 1M tokens, with maximum input of 991,808 (983,616 with thinking) and output capped at 131,072 (262,144 with thinking).

Should you switch to it? If your workload is text-heavy with long outputs — drafting, coding, document work — it is the cheapest flagship per output token in the comparison table. Run your own fifty-task evaluation against your current model before you move anything production; the price advantage survives most evaluations, the quality advantage is provider-reported so far.