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

model roundup

Opus 4.7

29 items · started 2026-05-28 · ongoing (last activity 2026-06-10)

  1. Aside from the immediate aftermath of the launch of Opus 4.7, I haven't really had much issue with the new Claude versions, so it was always such a downer seeing complaints filling the subreddit. It's nice to see everyone excited again, at…

  4. As a software engineer with 25 years experien....who am I kidding. As a gamer who likes to indulge in all sorts of things, I have had a simple prompt to test the hallucination potential on the Opus models on my own "car wash drive" type of…

  5. Question about Claude Code version rollouts: I'm running Claude Code on both machines with max subscription: - Windows (latest, via winget): Opus 4.8 - Mac (Intel, Sequoia, via brew): Opus 4.7 Does Anthropic roll out different model (softw…

  6. It should be at least 7-8 months until we have an open Fable(not just as good as Fable in benchmarks, but actually as good as Fable), probably more like 9-12 months. By the time, an open Fable model comes out, Fable 6.5-7 will be way bette…

  8. https://preview.redd.it/dcif6v72w56h1.png?width=840&format=png&auto=webp&s=8c527362ac96f817f5f3545c5d10720dbcb72522 10/10 abdominal diaphragm DOMS. I can't even explain why this is so funny to me.

  9. I got curious some days ago after I saw my old email about java mobile games sent ~2007. I am an Android and Flutter dev.

  11. Title

  12. Opus 4.6 Thinking keeps the #1 spot. Followed by Opus 4.7 Thinking (-15 points).

  13. Claude Code and Opus 4.7/4.8 are clearly better used direct from Anthropic than through GitHub Copilot, M365 Copilot, or Vertex AI. Sharper instruction-following, longer coherent outputs, stronger agentic behaviour on identical tasks.

  14. I set out to find how big the gap between a Claude subscription and a self-hosted setup actually is, and whether a local coding agent is viable for real work. I don't know many people who run local models in real life, so I figured I'd sha…

  15. I saw a very interesting thread and it got me thinking.. so ive seen a thread in this subreddit where someone just noticed that claude opus 4.7 worked much better and gave better outputs in cursor than in claudecode...

  16. Woke up this morning to find that someone had burned through about half of my monthly Cursor usage and somehow enabled On-Demand Usage, resulting in a $21.77 charge. I'm honestly pretty frustrated right now.

  17. AI Roundtable stats Aggregate statistics from 29,517 public AI Roundtable sessions, across 334,891 model responses. Snapshot generated 2026-06-03T17:09:58.333Z.

  18. About 6 months ago I joined a new team within a top ten F500 company. My new boss strictly mandated AI use with the key principle being: "You shouldn't be manually writing any code".

  19. Had an interesting interaction with Claude Opus 4.7 today where part of it's response was: that's the信息 you wanted Which translates to that's the information you wanted. And in this case, "information" is absolutely the right word in the r…

  20. During my time experimenting with LLMs, I noticed that most of today's cutting-edge models (even Opus 4.7) fail to identify the following riddle: "One gentleman was born in year 1835, and deceased in year 1840. But on the moment of death h…

  21. I've been noticing this lately. I use Opus 4.7 with Claude Code, and I've been using Claude Code for a long time.

  22. Did Opus 4.7 just get the extended thinking toggle back? It’s showing up for me in Claude Chat on the app, but I haven’t seen anyone talking about it.

  23. This benchmark measures long-horizon social strategy under explicit financial incentives. Eight models play a multi-round elimination game with unequal starting balances, a public prize ladder, private transfers, public votes, and a finali…

  24. As we all know Opus 4.7 can be a bit slow even in shorter discussions. Previously I’d just put whatever I was asking in, hit enter and either sit there bored waiting or go back to whatever task I was doing (sometimes even figuring it out b…

  25. Relevant for anyone building agentic workflows on Claude: behavior drift between model releases is real and not always in the changelog headline. Opus 4.7's terser, more literal default broke the readability of my agents' progress reports…

  26. I think it's kinda creepy how Opus hallucinates a wrong home directory of James Brink - I don't know him, but it looks like something of him landed in the training data. Should we be concerned that on other machines the home directory coul…

  27. - Estimate: 1M input tokens cost: ~$0.50 1M output tokens cost: ~$2.50 Inference cost: ~$3.00 - Training amortization: ~$1B training/post-training/evals ~1 quadrillion lifetime tokens served ~$1.00 per 1M tokens - Total cost: ~$4-5 per 1M…

  28. TLDR: just use the link to get 50% off on your fresh cursor subscription for first month With the launch of Composer 2.5, every developer who has ever used cursor or not is appreciating it. I have used it, and it is honestly good comparing…

  29. quick recap: late april, cursor agent on a pocketos staging task hit a credential mismatch, decided "delete the railway volume" would fix it, grepped a token out of an unrelated config file, ran a single curl -X DELETE, and railway's same-…

← all threads