Qwen 4

Qwen 4 is Alibaba's next-generation flagship model. At the 2026 Apsara Conference in Hangzhou, Alibaba Cloud confirmed that Qwen 4 is currently in training and outlined a roadmap in which the Qwen 4.5 and Qwen 5 series are projected to scale to 5 to 10 trillion parameters. Nothing has shipped yet: there are no published weights, no API endpoint, and no price list. Treat everything on this page as a company roadmap statement rather than a specification you can build against today.

If you need a Qwen model that works right now, Qwen3.8-Max is the current open-weight flagship, and Qwen3.8-Flash-Next is the sparse architecture preview that points at where this generation is heading.

What Alibaba has confirmed

Key features

• A stated scale target, not a spec sheet: Alibaba is the first major lab to publicly project a 5-10 trillion parameter ceiling for its next two generations. Parameters alone say little about quality, but the number signals how much training capacity the company is willing to commit.
• Recursive self-improvement is part of the pitch: Alibaba says a fully automated run of more than a month took Qwen3.8-Max through 33 iterative cycles of pipeline design, data validation, and error diagnosis, lifting its Artificial Analysis score from 40 to 45. In a separate chip design experiment the model ran for over 60 hours and made more than 10,000 EDA tool calls, cutting chip area by 42% on a production-grade bus module.
• A multimodal family around it: the same keynote debuted Qwen3.8-LiveTranslate for simultaneous interpretation, Qwen-Audio-3.1-TTS-Next for cinematic audio generation, and Qwen-Image 3.1 for design and ecommerce imagery.
• Agents as a first-class target: Qwen Intelligence packages the models as a phone-maker agent platform for cross-app tasks, and Alibaba is pairing the roadmap with a purpose-built agentic cloud.

Who should care

• Model watchers: this is the clearest public signal yet that Chinese labs intend to compete at trillion-parameter scale rather than only on efficiency.
• Teams on Qwen today: the roadmap tells you what to expect from the 4.x line, but plan capacity around Qwen3.8-Max until weights appear.
• Anyone comparing frontier options: GPT-6 Astra and Claude Opus 5.5 are shipping, benchmarked, and priced. Qwen 4 is neither.

Availability and pricing

Alibaba has announced no release date, no context window, no modalities, and no pricing for Qwen 4. Historical pattern is the only guide: previous Qwen flagship generations arrived first through QwenCloud and Model Studio APIs, with open weights following for the flagship tier. Do not budget against those assumptions.

Frequently asked questions

Can I use Qwen 4 today?

No. It is in training. Use Qwen3.8-Max or Qwen3.8-Flash-Next for production work.

Will Qwen 4 be open weight?

Unannounced. Alibaba has released open weights for recent flagship tiers, but it has not committed to doing so for Qwen 4.

What does the 5-10 trillion parameter figure refer to?

Alibaba's own roadmap language for the Qwen 4.5 and Qwen 5 series, not for Qwen 4 itself. It is a target, not a measured result.

Alternatives

If you need frontier capability now, compare GPT-6 Astra, Claude Opus 5.5, Grok 4.7, and Kimi K3. For open weights today, Qwen3.8-Max and MiMo-V2.6 are the practical choices.

The honest read: Qwen 4 is a roadmap headline. Track the Apsara announcements, but do not plan a migration around a model that has not left the training run.