Qwen (通义千问) Review 2026

Qwen, from Alibaba Cloud, is the answer to the question "what if frontier AI cost almost nothing?" Its Qwen3.7-Max model ties Claude Opus 4.7 on the Artificial Analysis Intelligence Index while its API costs a fraction as much, 3 to 60 times cheaper depending on the tier. The weights are fully open, so you can self-host and fine-tune, and the 1M token context comes with no long-context surcharge, a big deal for anyone processing long documents.

Its trick card is multimodal: text, vision, and up to two hours of video input in one prompt, something DeepSeek does not offer. That alone settles the comparison for many builders. DeepSeek still leads on pure coding benchmarks, but for multimodal work Qwen is the pick.

The caveats are honest ones: content moderation follows Chinese regulations, documentation is Chinese-first, and the model lineup (3.5, 3.7, 3.8, Max, Plus, Flash, Coder) is genuinely confusing at first. The $50 a month coding plan gets you 90K requests across seven models. Pricing checked August 28, 2026.

The August 26 release is the one to watch. Qwen3.8-Flash is a sparse model, roughly 125B total parameters with about 6B active per token, priced at $0.15 per million input tokens, about three times cheaper than DeepSeek-V4-Flash, and Alibaba claims roughly a 90% drop in training cost versus the prior generation. The weights are open-source. The same release shipped Qwen Office at ¥98 per month, Alibaba's office agent built around a Standard Mode that the company says runs single tasks about twice as fast at roughly 75% fewer tokens. For the full picture of that launch week, read our Chinese AI office war recap.