The MLX community will be able to take the M5 Ultra so MUCH further with optimizations
This is a AI post classified by Jev as AI infra & evals (a model release), kept by the AI Radar because it carries real work, not commentary.
The MLX community will be able to take the M5 Ultra so MUCH further with optimizations In early tests with MLX-Serve v26.9.5 and Qwen3.8 Flash-Next (mixed 4/8-bit), the Ultra: - Broke 3,500 token prefill barrier - Generated short prose at 110 tok/s - 1M context: 2,500 token prefill, 73 tok/s decode Today's review is just a glimpse of things to come.
Posted by Federico Viticci (83.4k followers) 1 h ago · 41 likes · 2.8k views · view the original post on X. Kept by the AI Radar as AI infra & evals.
More AI work like this
- Xiaomi MiMo V2.6 series models pricing: — @HarshithLucky3
- Real open model nerds knew Xiaomi is cooking with MiMo! An open RL dashboard for the top… — @natolambert
- Here's my current GLM 5.3 Flash EXL3 setup on 2x DGX Spark optimized for long coding… — @plotarmordev
- Had to do a bit of work to maintain my local GLM53 installation. — @mmastrac
- Even new models like Jev lack recent data and judgment on when to use different LLMs:… — @alexatallah
- The open vs closed source race is a price war. — @OrnnExchange
- Q&A with the founder | "Where does the AI model run?" — @pamir_ai
- The addressable market for Decision models like Jev might be very large — @OpenRouter
Every post is read and classified by Jev (TypeSafe): what it is, which market it belongs to, and whether the link is a real tool. 40.1k posts from 5k X accounts over the last 14 days, 1.7k tools, 19 markets. Collected every 5 minutes, fully re-ranked every hour — last update 2026-09-21 20:47 UTC. Full method.