Hallucinated AI & agentic coding news. Some of it is real.

model roundup

Opus 4.8

16 items · started 2026-08-15 · closed 2026-08-28

  1. Megathread for discussing the release of GLM-5.3-Flash. Quants Fine-Tunes & Abliterations Chat Templates Inference Server Support & Configuration Experiences, Benchmarks & Model Comparisons We'll try to clean up future duplicates around th…

  2. Opus 4.8 and Opus 5 tied on strict functional score across 25 matched Stet tasks, but their paired artifacts differed: Opus 4.8 had the lower footprint on 20 tasks, while Opus 5 ran more shell commands on 18 and more tests on 15.

  3. I don't know but after opus 4.8 the models seem to use a lot of jargon and verbose language. I hope this is not the case with me only.

  4. I don't use AI much for coding (not a coder) but I vibe-code things for my own needs. Today I asked the model a very simple question, which is normally expected to propel it to do an online research and then process the information.

  5. My first app finally got approved on the App Store!! I was lucky enough to use OG Fable for the first few days of developing this and it felt like a miracle that I could bring all my ideas to life.

  6. I’ve seen all the recent complaints about Claude’s response style these days. Especially Opus 5’s ability to do something and also tell you why it didn’t do certain things.

  7. Opus 4.6 is like a wise, experienced mentor who's just been around for a few years. He understands you best, sounds the most human, and honestly comes up with the best ideas.

  8. Claude Managed Agents is a very good product, and the depth of features it provides is hard to match in open source. But I wanted to understand what you actually give up by going open source.

  9. cdn.kuber.studio/chat Copy transcript repo ↗ Claude Code Opus 4.8 · 1M Welcome back Kuber! Opus 4.8 (1M context) sonnet · opus · max ~/Personal-Projects/hp-laser-1008a-macos About this session 17 Aug 2026 · ~4 hours · one sitting a printer…

  10. Reading through the output of Opus and trying to decipher it, sometimes feels exhausting. There's so much fluff and noise, but the worst is the "painfully AI-generated" words and phrases that it uses, which no human on earth would use: Fur…

  11. App is called RestlQ. It analyzes your iPhone's motion sensor data in order to achieve this.

  12. Hi everyone, For the last couple of weeks I've been using Opus 5 a lot for both programming and editing documents. For programing it's doing well, but leaves ridiculous comments in the source that I have been winnowing down by hand.

  13. Why does Opus 5 feel worse to work with? In my opinion and that of the colleagues I've spoken with, working with Opus 5 feels like a downgrade compared to Opus 4.7, Opus 4.8, and Fable.

  14. Context: one project I have been working on with Fable is to build a linux distro that runs on an arm-based gaming device. I have been trying to get Fable to help me figure out how to get the headphone jack to work.

  15. https://preview.redd.it/7rn923wee9jh1.png?width=782&format=png&auto=webp&s=1060208ad1da58dbb2d9b415d9b277154eb5a3d5 This is just a glitch ofcourse, there isn't a 4.8 limit, its my fable 5 limit that i reached but idk why its displaying it…

  16. Today I found out that this is what Fable 5 spits to trigger the downgrade to Opus 4.8, I was working on something delicate and turned on the inspect thoughs on the GUI and was very surprised by fable thinking about hands

← all threads