Day 2 of making DeepSeek V4.1 Flash fast on my 4x 3090. I still cannot believe what is…
This is a dev post classified by AGK Intelligence as AI dev tools (a tool drop), kept by the Dev Radar because it carries real work, not commentary.
Day 2 of making DeepSeek V4.1 Flash fast on my 4x 3090. I still cannot believe what is happening. 3 quants: - Maya-S by @PeasantSmith - @UnslothAI UD-IQ3_S - SAGE 1.62 bpw by @ViC305 Maya-S on replays of my agent sessions, against my old build. Full run: 1159 s to 720 s. First token: 4.8 s to 1.7 s. UD is running right now. As a bonus there will be a few PRs to the Maya project, I hope they will help a lot of people. I did not expect this result. Tomorrow a thread with all the results and charts.
Posted by Alexey Fateev (9.7k followers) 3 h ago · 31 likes · 1.8k views · view the original post on X. Kept by the Dev Radar as AI dev tools.
More dev work like this
- A BIG update is here! Qwen3.8 Flash Next now runs on a 8GB graphics card at 20+… — @coldniko
- Say hello to Deepseek Coder in Strata! 👋 — @coldniko
- Qwen3.5-4B-GLB-3D-Coderで作った3Dモデルを使ってRPGを作った。クセは強いけどかわいい。 — @kis
- Meet Qwen3.8-27B-USS-Prism-q4-hi-mlx: a 27B image-text-to-text model that runs on MLX.… — @HuggingModels
- Factory's @luke_alvoeiro explains why AI coding agents are wasting massive amounts of… — @MTSlive
- you can just "vibe train" models now. i built a bridge simulator (with real physics) and… — @kylejeong
- Factory's @luke_alvoeiro explains how Droid beat Anthropic's native Claude Code harness… — @MTSlive
- These are the 3D builds that used to break, now they don't. — @ItsmeAjayKV
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.4k posts from 5k X accounts over the last 21 days, 9.1k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-03 00:07 UTC. Full method.