Dagens Vibes — 18. august 2026

This is unreal. Qwen 3.8 27b running locally on my RTX 5090 BEATS Opus 4.8 in my benchmarks (funny enough slightly loses to Opus 4.6 lol) Speeds are just as good if not sometimes MUCH better. Got up to 200 tk/s in some tests We live in a new world. Unlimited, free super intelligence on our desks No internet, subscription, or APIs required Just 6 months ago you needed 500gb of VRAM to get a quarter of this performance Now I'm getting Opus 4.8 level intelligence on a single RTX 5090. Don't even need a Spark I now have the model powering a Hermes agent that is doing work for me 24/7 around the clock for free I have ZERO doubt that by this time next year we will have models more powerful than Fable running on Mac Minis. ZERO doubt. If you listened to me earlier this year when I demanded you start buying hardware, you are absolutely eating right now. The future isn't around the corner. It's here.

https://x.com/AlexFinn/status/2089455600338116980

a day ago i open sourced a repo for qwen 3.8 27b dense with two speed rows in it. it now holds numbers measured by 28 people on their own hardware, pascal to blackwell. if you run local ai on consumer hardware and you are not watching this repo, you are missing out. qwen 3.8 27b dense, baseline → with the mtp flag, tok/s: workstation > rtx pro 6000 blackwell 96gb: 63.8 → 91.5, max-q 45.7 → 97.1 > rtx a6000 48gb: 26.7 → 52.5 rtx 5090 desktop, ten configs > baselines 61-77 → 109-180 with the flag, crown 179.7 at ud-q4 262k rtx 5090 mobile > 36.7 → 50.9 rtx 4090 > 47.7 → 76.3, and 36.1 → 74.8 at 160k rtx 3090 and 3090 ti > 31 → 41 day one, 41 → 63 on a fresh build, 40 → 62 turboquant fork > overclocked: 44 → 80 at q4, 52 → 86 at q2 > 3090 ti: 42 → 61 at both 128k and 256k radeon > rx 7900 xtx: 30.7 → 43.9 windows, 28.8 → 70.7 vulkan linux > rx 7900 gre, packed 16gb: 28.7 → 47.8 under live agent traffic > radeon pro r9700: 27 → 43.3 > radeon 9060 xt: 15.2 → 28.7 apu and igpu > ryzen ai max 395: 11.5 → 23.7 windows, 10.5 → 21.4 rocm > radeon 890m igpu: 2.7 → 5.7 multi-gpu > 3x 3090 tensor parallel: 49 → 81 > 2x 5060 ti: 37 → 66 and 38 → 76, tensor split doing the heavy lifting > 2x rx 9070: 22.1 → 41.6 > 3x 3060: 17.3 → 24.5 39 configurations. 31 prs merged. 7 tuning rules discovered by contributors, not by me. 36 forks. one free flag. if you have numbers, open a pr, the next person's perfect config depends on your row. and if you own hardware already in the table, run the probe and VERIFY the row, replication is the half of science nobody does. the table takes both.

https://x.com/sudoingX/status/2089385342747168839

Claude Code can design now. The new /design skill (research preview) brings Claude Design's artboard workflow into the CLI and Desktop, built on artifacts. Run /design to get editable artboards for your UI — pick one, tweak it, then have Claude implement it. https://t.co/uttl4F1G0e

https://x.com/ClaudeDevs/status/2089471692762673408

Good moment today to remind everyone to stay frosty with AI. I was working on making a server stoppable via the CLI, and Codex decided the best path forward was to implement an unauthenticated network API to every server to shut itself down. Excellent work. I responded with "I don't know about that." (I usually give more direct feedback but this was so absurd) and it thought for like 20 seconds and responded "Server stopping should not be part of an unauthenticated network API." Yes. lol. By the way, the result I was looking for was that we automatically start a locally discovery Unix-socket bound server so I wanted it to discover that (using prior mechanisms already made) and gracefully stop that. The driver is absolutely accountable here (me). My lazy prompting led AI to think I wanted to be able to stop ANY configured server I was talking to. Shame on me for thinking it was obvious here. But, I also reviewed it and found this idiocy, so, that's where good AI drivers come in. Also me.

https://x.com/mitchellh/status/2089446702285758559

I haven't been at my desk for 4 days, I come back to see this. An AI set up a fake mosquito floating around and a calibration routine to test how well my stereoscopic camera would be at detecting mosquitoes Just needed to point it at a monitor https://t.co/fjaXgtLhPn

https://x.com/yacineMTB/status/2089394881760264538