Qwen 3.6 35B-A3B lands in Puter
- Puter said this week that developers can now call Qwen 3.6 models in Puter.js, including Alibaba’s open-weight Qwen3.6-35B-A3B, without API keys. - The model uses 35 billion total parameters but activates 3 billion per token, and Puter highlighted 73.4% on SWE-bench Verified. - The release folds a newly open Qwen coding model into browser-accessible tooling for cheaper testing. (developer.puter.com)
A large language model is software that predicts the next token, or chunk of text, and developers use stronger versions to write and debug code. Puter said this week it added Alibaba’s Qwen 3.6 lineup to Puter.js, including Qwen3.6-35B-A3B. (developer.puter.com) Puter’s April 27 post said the models are available “free for developers” in Puter.js and do not require API keys. The company listed Qwen3.6 Max Preview, Qwen3.6-27B and Qwen3.6-35B-A3B as available options. (developer.puter.com) Qwen3.6-35B-A3B is a mixture-of-experts model, a design that works like a team where only a few specialists answer each request. Qwen said the model has 35 billion total parameters but only 3 billion active parameters per token. (qwen.ai) (huggingface.co) Qwen released Qwen3.6-35B-A3B as open weights on April 14, 2026, and said it is licensed under Apache 2.0. The company said the model scored 73.4% on SWE-bench Verified and 51.5% on Terminal-Bench 2.0. (qwen.ai) (alibabacloud.com) Puter’s pitch is access, not a new model release. Its blog framed the update as a way to use Qwen 3.6 through a JavaScript library in the browser or app code without separately provisioning cloud credentials. (developer.puter.com) That matters because Qwen’s own release positioned 35B-A3B as an “agentic coding” model for multi-step software tasks, repository analysis and tool use. Puter’s examples showed codebase tracing and PDF-to-database scripting prompts aimed at the same developer workflow. (qwen.ai) (developer.puter.com) Qwen’s broader 3.6 family arrived in stages over April, with Qwen3.6-Plus first and open-weight variants following later. Puter’s post said its integration also includes Max Preview, which it described as topping six coding benchmarks. (developer.puter.com) (github.com) The practical appeal of 35B-A3B is that the active slice is much smaller than the headline parameter count. Qwen and third-party hardware guides both describe that architecture as the reason the model can be deployed more efficiently than a dense model of similar size. (qwen.ai) (willitrunai.com) For developers, the update means a newly released open Qwen coding model now sits behind a browser-friendly JavaScript call. Puter is selling convenience; Qwen is supplying the weights and benchmark claims. (developer.puter.com) (qwen.ai)
