Alibaba releases Qwen3.5, extending its open-weight line
Originally announced by Qwen . We link the primary source so you can read it for yourself.
Alibaba’s Qwen team has released Qwen3.5, the next step in one of the most active open-weight model lines. The release continues the family’s Apache 2.0 licensing, which is much of why Qwen models have become a default choice for people who want a capable model without commercial-use conditions to reason about.
What was released
Qwen3.5 follows the Qwen3 line, which introduced the family’s hybrid thinking mode and its 128k context window. Qwen positions the update as an across-the-board improvement, and the announcement carries the usual set of benchmark comparisons. In keeping with our approach to any figures published by a model’s developer, we treat those as claims to be confirmed against independent evaluation rather than settled results.
Why it matters
The significance here is less any single number and more the cadence. Qwen has shipped frequently and under a permissive licence, which has kept it near the front of the open-weight field and made it a reliable base for local setups. For people running models on their own hardware, a steady stream of Apache 2.0 releases across a range of sizes is exactly what makes a family worth committing to.
The context
This sits within a broader pattern in 2026 of open-weight labs, several of them Chinese, releasing quickly and licensing permissively. It follows the Qwen3 series and precedes further updates in the same line. Whether Qwen3.5 holds its position will depend on independent testing and on how the rest of the field responds, rather than on launch-day benchmarks.
We will add Qwen3.5 entries to the model catalogue once we have verified their specifications and memory requirements against primary sources.