Qwen Chat
Alibaba's chatbot on top of the largest, most complete model family in AI — spanning text, code, vision, audio and video under one roof.
What is Qwen Chat?
Qwen (通义千问) is Alibaba Cloud's answer to nearly every category of AI model at once. Rather than ship one chatbot and call it done, Alibaba has built out more than 145 model variants spanning text, code, vision, audio, translation and video generation, all accessible through the same Qwen Chat interface or developer API. For years, the defining trait of the whole family was openness: most Qwen models ship with their weights publicly released under Apache 2.0, which is part of why Qwen has become one of the most self-hosted model families among developers who want a capable model without a per-token bill.
That's shifting, though. The newest flagship generation, the Qwen 3.7 series, launched in 2026 as API-only, with no open weights released alongside it — a real change in direction for a company that built its reputation on giving models away. The flagship, Qwen3.7-Max, can now run autonomous agent sessions for up to 35 hours, chaining more than a thousand sequential tool calls without falling apart, and Alibaba reports it has the lowest hallucination rate among frontier models it tracks. Meanwhile, the open-weight side of the family, led by Qwen3.5 and Qwen3-Coder, remains genuinely excellent and is still where most developers actually deploy Qwen in production.
Traffic and hallucination-rate figures are Alibaba's own reported numbers as summarized in industry coverage through mid-2026, not independently audited. Model availability (open vs. closed weights) is changing quickly — check Qwen's official model page for the current status of any specific version.
Key features
One ecosystem, every modality
Text, code, image, audio and video generation all live under the same Qwen umbrella.
Thinking mode
A toggleable extended-reasoning chain, visible directly in the Qwen Chat interface.
Long autonomous sessions
The flagship model can run for up to 35 hours, chaining over 1,000 tool calls on a single task.
Strong coding models
Dedicated Coder variants rank among the best open models available for real-world programming.
Exceptional multilingual reach
Supports 201 languages and dialects, well beyond what most Western competitors cover.
Open-weight options
Several strong models remain freely downloadable and self-hostable under Apache 2.0.
Pricing
Free chat app
- Access to Qwen's consumer chat interface
- Thinking mode & multimodal input included
- No monthly plan to compare or upgrade
API — open-weight tier
- Covers Qwen3.5 & Qwen3-Coder class models
- Same weights available to self-host for free
- Some providers add a free tier for prototyping
API — Qwen3.7-Max
- Current flagship, closed weights
- Long-running autonomous agent sessions
- Roughly a fifth of Claude Opus 4.8's per-token cost
Alibaba discontinued its free OAuth API tier on April 15, 2026 — free hosted access now mostly comes through third-party providers like Groq rather than Qwen directly. Exact API rates vary by model and provider; figures shown are approximate as of July 2026.
Available models
Integrations & platforms
Pros, cons & best for
Pros
- Broadest model family of any lab, spanning nearly every modality
- Genuinely excellent open-weight options for self-hosting
- Class-leading multilingual support
Cons
- Newest flagship models are now closed-weight, reversing past openness
- Free API tier was discontinued in April 2026
- Reported caching inefficiency can inflate real-world API costs
Best for
- Developers wanting a strong, self-hostable open-weight model
- Multilingual products, especially outside English-speaking markets
- Long-running autonomous coding or agent workflows
Take a look inside
Alternatives
For other strong open-weight options with a different focus:
Our verdict
Qwen's scale is genuinely hard to match — no other lab covers this many modalities this well, and its open-weight models remain some of the best available to self-host, full stop. The direction of travel is worth watching, though: the shift of the newest flagship to closed weights breaks from the openness that built Qwen's reputation, and the free API tier that used to make experimentation easy is gone. If you want a capable, self-hostable model with unusually strong multilingual reach, Qwen remains one of the best choices around; if you need the absolute frontier model and don't mind paying, the Max tier holds its own against much pricier Western competitors.
FAQ
Is Qwen Chat free?
Yes, the consumer chat app is free to use. Alibaba charges separately for API access to Qwen's models, billed per token.
Are Qwen's models open-source?
Many are — most of the Qwen3 and Qwen3.5 family is released under Apache 2.0 and can be self-hosted for free. The newest flagship generation, Qwen3.7, is API-only with no open weights released.
Is there still a free API tier?
Alibaba's free OAuth tier for the Qwen API was discontinued on April 15, 2026. Free hosted access to some Qwen models is still available through third-party providers like Groq.
What is Qwen best known for?
Breadth and price-to-performance: it covers more modalities than most competitors and its open-weight models are widely regarded as some of the strongest available for coding and general use at a fraction of the cost of closed frontier models.
Why does Qwen have so much traffic from Russian-speaking users?
Reported figures suggest roughly 30% of Qwen's platform traffic comes from Russian-speaking users, likely reflecting strong multilingual support and more limited access to some Western AI products in that market.
Can Qwen run long, autonomous tasks?
Yes, the flagship Qwen3.7-Max can run agentic sessions for up to 35 hours, chaining over a thousand sequential tool calls on a single task.