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

model roundup

Opus 4.6

13 items · started 2026-07-24 · closed 2026-08-07

  1. Finally had the privilege of fable spinning up 70 subs agents My max 5x plan stood no chance, didn’t even finish the first prompt and only lasted 20 minutes into the 5 hour session Granted it was doing a heavy research pass for my hat comp…

  2. Dear Anthropic, can we please have thought traces back? I can't verify whether or not the LLM is arriving at the conclusion from cheating, or if it's fudging or making stuff up.

  3. Suggest me one. My priority is to do advance audit, bug finding and resolve them in my Flutter code.

  4. TLDR: Be careful with the use of subagents by actively limiting the number that can be created and don't allow them to spawn their own. Today, I ran into an issue with a prompt that I run frequently with Opus 4.6, 4.7, 4.8 with subagents t…

  5. I'm in Claude Code on Opus 4.6 at 25% of my session limit, 44% of my weekly limit, and it just charged me $4.60 in usage credits for a small prompt and won't work if I turn usage credits off. Wtf?

  6. I'm wondering which model is currently the best for content writing that follows large instructions and produces natural tone, that is easy to read. My experience says it's Opus 4.6, but then it does not follow all the instructions.

  7. I still can't bring myself to move away from Opus 4.6. I feel like it's more than enough to use, and the UI is good enough that I don't see any real need to upgrade.

  8. I gave eight model and effort configurations the same prompt: design when a manager should use zero, one, or several AI advisers for an important decision without creating a permanent committee. This was one judged strategy sample, not a g…

  9. I'm interested in learning from you bluds faring and building production grade projects. Excluding plugins and skills from 3rd party sources.

  10. I like using Claude as a talking partner to talk about my writing ideas and DND campaign planning with, simple stuff, mostly really character focus as I end up finding mapping out exact mindsets really fun. I have a pro subscription for it.

  11. I was having Claude review and revise a document using Opus 4.6 Medium and a skill we created. I noticed that it accidentally deleted chunks of text without noticing.

  12. Hi All!. :) Over the last few weeks, I’ve been experimenting with how far AI-assisted development can go beyond the usual web applications and automation scripts.

  13. Opus 5 scores just below Fable 5 (1.3 percentage points lower), but vastly outperforms Opus 4.6, 4.7 and 4.8, and all other models tested. > SimpleBench includes over 200 multiple-choice questions covering spatio-temporal reasoning, social…

← all threads