Model Releases

Guys, I am absolutely astounded. The Qwen 3.6 27b is like a jump to Qwen 4 from Qwen 27B 3.5. I just did a full suite of front end design te…

Guys, I am absolutely astounded. The Qwen 3.6 27b is like a jump to Qwen 4 from Qwen 27B 3.5. I just did a full suite of front end design tests and agentic benchmarks, made entirely by it. VERDICT: Th

DGX agentx-post
model-releasesclem-delangue--x

Guys, I am absolutely astounded. The Qwen 3.6 27b is like a jump to Qwen 4 from Qwen 27B 3.5. I just did a full suite of front end design tests and agentic benchmarks, made entirely by it. VERDICT: They're so much better than I thought they'd be, like I'm completely astounded. I feel like I'm getting front ends that Gemini 3 Pro was making last year, which shook everybody on release then, and it was a SOTA model for a time. And this is running on a single RTX 5090 on my own machine. I could iterate 200 versions of every design and pick the best. The tokens are FREE! It does like to think a lot, but it genuinely is still astoundingly good with it turned off. Read the full write-up for more insight on that! This is the "bye bye subscription era" as my friend @DJLougen so eloquently put it. You can open and interact with all of the benchmarks right now on my HF space as well as read the entire write-up from my tests! https://huggingface.co/spaces/KyleHessling1/qwen36-eval

Source: Clem Delangue (X) | 2026-04-22

Loading related sources…