Qwen: the open-source model juggernaut, explained

Qwen is Alibaba's family of AI models, most released as open-weight under the permissive Apache 2.0 licence — free to download, run, fine-tune and use commercially. The family spans text, code, vision and image generation, from 0.6B to 235B parameters. That openness made it the most-downloaded open AI family in the world: by April 2026, MySummit reports, over half of all open-source AI downloads worldwide, approaching a billion total, across 433 open models on Hugging Face.

On 20 April 2026 Alibaba shipped the Qwen3.6 generation. The headline finding: Qwen3.6-27B — fully open, Apache 2.0, runs on a single consumer GPU — reportedly outperforms a 397B predecessor on agentic coding benchmarks (Alibaba's numbers). Efficient architecture (Mixture-of-Experts on the larger 35B-A3B variant) is the mechanism: better engineering, not brute-force scale.

What to watch. The open models remain genuinely capable and freely available — this is not a retreat from openness, it is the standard open-weight monetisation pattern. The durable question: does the gap between open and closed tiers widen over successive Qwen generations? If the open tier stays competitive, the Apache 2.0 promise holds. If it becomes a demo funnel, the story changes.