The town made with the Qwen 3.8 27B model running DeepSeek Harness looks pretty good too!
Especially the light and shadow effects, switching between far and near lenses, it feels like it can be played! The total cost is only 211k tokens, and the entire process is completed locally. Thanks to @loktar00 for bringing us the test!
Another side by side comparison of Qwen 3.8 Max vs. QoDuck (presumably some Qwen 4.0 variation).
Both using the same prompt and thinking level. Qwen 3.8 Max took 1 hour and 55 minutes with 137 iterations in one shot. QoDuck took 29 minutes and 30 seconds with 99 iterations in
Qwen 3.8 27b NVFP4 inside TensorFold native CLI on Windows 11 cooking on a nvidia RTX 5090.
🤯 Single-stream decode was ~220 tps for prose. Prefill was ~12,200 tps C=4 was over 500 tps decode. Will be in the next engine release 👌🏼
20M Tokens FREE Every Day on APInex_AI🆓
The promo continues. Use your 20M daily allowance on top models: 1️⃣ DeepSeek V4.1 Flash 1017 2️⃣ Kimi K3 3️⃣ Qwen 3.8 MAX 4️⃣ GLM 5.3 Flash 5️⃣ Gemini 3.8 Flash ⚡️ Works in Cursor, Claude Code, Hermes and OpenCode 👉
You have reached the end of the archive
All of qwen38