MTPLX just made Qwen 3.8 run almost twice as fast on a Mac, with zero loss in output quality, and the math behind it actually proves it.
This breakdown covers what MTPLX by Youssofal_ is, how it uses native multi token prediction to speed up local models on Apple Silicon
A summary video of AI news covering Meridian and Qwen 3.8 Omni has been released
A summary video of AI news covering Meridian and Qwen 3.8 Omni has been released, from topics of changing the camera position of captured images to reports of intrusion into OpenAI, 41 minutes 41 minutes of AI uploaded by Bilibili's channel "设计乡之道" on 2026-09-24
And now I present to you
A "beautiful rendition" of world of warcraft in three.js by Qwen 3.8 27b... (having removed the conical trees) While using a custom vibe coded harness and, pointlessly, djev.
On my Strix Halo PC, Qwen 3.8 Flash-Next on Halogen passed 66/72 coding jobs and generated at 38.0 tok/s versus 13.9 in my earlier setup.
Different quant, backend and drafting make this a configuration comparison, not an engine-only test. Runs:
You have reached the end of the archive
All of qwen38