Last updated: 11/22/2026Last verified: 2026-11-15
Alibaba’s Qwen assistant: the 千问 app for AI search, slides, document analysis and image generation, plus Qwen APIs on 千问 AI 平台 with the Qwen3.8-Max and Qwen3.8-Flash flagship models. Qwen 4 is still in training; the only Qwen 4 model available is Qwen4-Max-Preview, a Token Plan-only preview that the docs say will be retired after November 25, 2026.
What is 通义千问?
通义千问 is Alibaba’s Qwen assistant, and the name now covers two products. The 千问 app is the consumer entry point: Alibaba’s product page describes it as the first place to use the latest and strongest Qwen models, with AI search, web-page summaries, AI PPT, AI image and video generation, meeting transcripts and document handling up to 500 pages. 千问 AI 平台 is the developer side, selling the same models per token through OpenAI- and Anthropic-compatible APIs and through Token Plan subscriptions. The current flagships are Qwen3.8-Max, a 2.4-trillion-parameter MoE, and Qwen3.8-Flash, both with a 1M-token context. Qwen 4 is still in training; the only Qwen 4 model on offer is Qwen4-Max-Preview, a Token Plan-only preview that the docs say runs from November 11 to November 25, 2026.
Key Features
- 千问 app: AI search, web-page summaries, AI PPT from files or prompts, AI image and video generation, and meeting transcripts
- Document handling up to 500 pages with summaries, translation and data analysis, plus photo-based question answering, grading and mistake explanations
- Qwen3.8-Max, a 2.4-trillion-parameter MoE, and Qwen3.8-Flash, both with a 1M-token context; Qwen3.8-Flash accepts text, images and video
- OpenAI- and Anthropic-compatible APIs on 千问 AI 平台, so clients such as Claude Code and Codex can point at Qwen models
- Qwen4-Max-Preview: 1M-token context, 393,216-token maximum output, image and video understanding, thinking on by default, on Token Plan only until November 25, 2026
Best For
Pricing
Alibaba’s product pages checked on November 15, 2026 publish features but no consumer price list for the 千问 app, so the app is charged as free to try with paid tiers unconfirmed here. On the developer side prices are per token: Qwen3.8-Max costs ¥12 per million input tokens and ¥36 per million output tokens, Qwen3.8-Flash ¥0.8 and ¥2.7, with separate cache, batch and cache-hit prices listed on each model page. Token Plan subscriptions bundle monthly Credits at ¥39, ¥79, ¥139 and ¥312 per month for 20,000, 45,000, 80,000 and 200,000 Credits, with limited-time new-user prices of ¥19.5 and ¥39.5 on the first two tiers until 2026-11-30 23:59:59 (UTC+8). Qwen4-Max-Preview has no per-token price: on Token Plan it consumes Credits at as low as 20% of the standard rate until November 25, 2026.
Pros & Cons
Pros
- One account covers a consumer app and a developer platform, so prompts and models can move between them
- Qwen3.8-Max and Qwen3.8-Flash are both published with a 1M-token context, and Qwen3.8-Flash also takes image and video input
- APIs are compatible with OpenAI Chat Completions and Anthropic Messages, which the docs list alongside Claude Code and Codex clients
- Per-token prices, cache prices and Token Plan quotas are all published, so cost can be estimated before adoption
- Qwen 4 can be tried early through Qwen4-Max-Preview without waiting for a general launch
Cons
- Qwen 4 has no published release date, parameter count or benchmark result, and Qwen4-Max-Preview retires after November 25, 2026
- Qwen4-Max-Preview is Token Plan only and cannot be reached with a pay-as-you-go key
- Alibaba’s product pages are Chinese-first and do not publish a consumer app price list
- Model catalogues, promotions and base URLs differ between the Chinese and international documentation sites, so the console region has to be matched
Alternatives
FAQ
Details
Platform
Features
- 千问 app: AI search, web-page summaries, AI PPT from files or prompts, AI image and video generation, and meeting transcripts
- Document handling up to 500 pages with summaries, translation and data analysis, plus photo-based question answering, grading and mistake explanations
- Flagship models Qwen3.8-Max, a 2.4-trillion-parameter MoE, and Qwen3.8-Flash, both with a 1M-token context; Qwen3.8-Flash accepts text, images and video
- OpenAI- and Anthropic-compatible APIs on 千问 AI 平台, so clients such as Claude Code and Codex can point at Qwen models
Languages
Known limitations
- Qwen 4 has no release date: Alibaba said on 2026-09-22 that it is in training, and the only Qwen 4 model available is qwen4-max-preview, which the docs list as Token Plan only and say will be retired after 2026-11-25 24:00 (UTC+8)
- QwenCloud’s Token Plan docs say Token Plan is sold only in the Singapore region and that its sk-sp- keys are separate from pay-as-you-go keys; the Chinese pages name their own regional base URLs instead
- API prices are per-token and differ by model: ¥12 input and ¥36 output per million tokens for Qwen3.8-Max, ¥0.8 and ¥2.7 for Qwen3.8-Flash on 千问 AI 平台 (checked 2026-11-15)
- The product pages are Chinese-first, the 千问 app price list is not published on the pages checked, and this listing did not run a conversation or an API call







