Actual test competition: The same 3D arcade question, 4 models have strong front faces, the details of DeepSeek V4.1 Flash are the hidden killer!
The same prompt, look at speed, cost and finished product quality: 1️⃣ DeepSeek V4.1 Flash｜Maximum details: 2 hours and 27 minutes｜$0.79 The slowest, but the internal shelves, machines, clocks, and display cabinets are laid out layer by layer, most like a convenience store/game arcade that you can walk into. It can be used after it is finished, not just a facade. 2️⃣…
Just tried running qwen 3.8 27B locally on a 5090 and the results really surprised me.
I only give one prompt: create an animation that shows the model's maximum capabilities, not using images or videos, but using everything else possible, as intuitive as possible.
Musk said that in the future, AI will directly produce binary data, which may be too conservative.
I just saw an interesting demo. The author clicks on "Drawboard" in a desktop that simulates Ubuntu. There is actually no such software in the computer. Qwen 3.8 27B directly generates HTML/CSS/JS on-site, and the browser runs immediately. After a few seconds, an app appears that allows you to draw, choose colors, and undo. Inference speed is close to 2000…
I wanted to test what Alibaba_Qwen Qwen 3.8 27B running locally on a single 5090 could do in Row-Bot.
I gave it just one prompt: "I want you to create an animation showing what the current local model can do. Testing it to the extreme. Do not use image generation or video
You have reached the end of the archive
All of qwen38