Thread

1. onomojo · 1786044657 · focus · HN ↗

   Any benchmark showing Opus 5 as the best just loses credibility for me. Anyone who's actually used Opus 5 daily knows what I'm talking about.

   1. copperx · 1786044937 · focus · HN ↗

      I'm dumbfounded to see Opus 5 making SO MANY mistakes in coding simple stuff. Most times, Fable 5 comes out to be cheaper because it nails so many things much quicker than Opus 5.

      1. garciasn · 1786046393 · focus · HN ↗

         I have Fable plan and Opus implement. I haven't had any major issues working this way; however, Opus does seem plain fucking stupid compared to what I experienced with Sonnet previously.

         1. aenis · 1786046909 · focus · HN ↗

            I do the same, and generally have good results, but it does stupid things with gusto.

            I'd open a blog with "weird things Opus did". Today it launched a swarm of cpu-hogging processes to test if the widget showing machine and I/O load is rendering nicely and correctly. The test went fine, but it was no longer able to kill those processes since they were really effectively hogging the CPU in various ways - being diligent, some of them were hogging CPU, some were murdering the SSD, some were pounding on the network adapters. Took me 30 mins to recover the machine to a working state without killing the meaningful, messy, in-flight sessions i had going on on other projects.

         2. petesergeant · 1786047409 · focus · HN ↗

            > however, Opus does seem plain fucking stupid

            Infuriatingly so, in a way I don't remember Opus 4.8 being, but maybe I've just been ruined by Fable 5.

            1. hbn · 1786048486 · focus · HN ↗

               I bought my first LLM subscription with Claude right before they gave access to Fable 5.

               I got so used to it, when they finally pulled access for me and I had to go back to Opus I felt like I was working with my hands tied.

               I finally know what those women with AI boyfriends felt like when their app updated and it won't dirty talk with them anymore.

            2. moffkalast · 1786049833 · focus · HN ↗

               Fable has spoiled us all.
      2. usef- · 1786047118 · focus · HN ↗

         Weird how different people's experiences are. If it's making simple mistakes something must be wrong in your setup/context I assume? It's been solid for me, beyond the usual LLMisms that all models have. But I keep context pretty minimal.

         1. cromka · 1786048157 · focus · HN ↗

            Statements like this typically come from working on the same setup and context using different models. I actually have that very experience now; I work on something security-adjacent so Fable often drops out, at which point Opus behaves like its lobotomized half-sibling. Pardon me the language, but I can't find a better example to be honest.
         2. nimonian · 1786048651 · focus · HN ↗

            Agreed. Opus 5 is doing just fine, slightly better than 4.8. It's personality is insufferable, but I find myself catching fewer problems at code review. It generally understands my conventions and isn't so eager to accrue tech debt.
   2. visarga · 1786045080 · focus · HN ↗

      Sent to solve one task, came back with half of it solved and 2 more problems.

      1. capnjazz · 1786047074 · focus · HN ↗

         "One thing worth your attention", "Two things worth knowing", "One thing to eyeball"

         1. greenchair · 1786049291 · focus · HN ↗

            This is driving me crazy. opus 4.8 did not do this to me not (at least during pre-5.0 timeframe). Feels like the new cycle is one step forward, two steps back.
   3. logicchains · 1786045247 · focus · HN ↗

      "As you requested, I've finished task X. Honestly, task X turned out to require task Y, which I haven't actually done. Task Y is the next step if you'd like to continue along this route."

      1. pornel · 1786047572 · focus · HN ↗

         [delayed]

         1. dr_dshiv · 1786051186 · focus · HN ↗

            Belt and braces all the way down
   4. cromka · 1786045321 · focus · HN ↗

      Agreed, it's extremely frustrating. It's the only model that actually makes me curse when talking to it, even knowing how counterproductive it is.

      1. moffkalast · 1786049610 · focus · HN ↗

         I'd certainly rank it at the very top of the want to kill yourself when using it benchmark. With weaker models you can sort of understand, they're trying their best and failing, but this thing just channels its immense inteligence into being as annoying as possible instead.

         1. msp26 · 1786051026 · focus · HN ↗

            yep matches my experience completely

            But even fable has the annoying tendency to invent new jargon and produce an incomprehensible soup of text.

      2. hungryhobbit · 1786049827 · focus · HN ↗

         The cursing thing blows my mind. "User is upset? Let's make decisions even faster (ie. more wrong) because clearly that's what they want!"

         It's a simple switch to make: cursing = try harder instead of cursing = stop trying. Is it really impossible to train Claude that way?

   5. bontaq · 1786045680 · focus · HN ↗

      It's an infuriating model
   6. nomel · 1786046867 · focus · HN ↗

      What's the clear best, that you see?

      1. petesergeant · 1786047072 · focus · HN ↗

         Fable 5
      2. drschwabe · 1786047678 · focus · HN ↗

         GPT 5.6 Sol
      3. kachnuv_ocasek · 1786049928 · focus · HN ↗

         GLM 5.2
   7. enraged_camel · 1786046935 · focus · HN ↗

      It's my daily driver. I like it and find it noticeably better than Opus 4.8.

      After I started reading complaints about Opus 5, I gave Fable the task of evaluating a bunch of code Opus 4.8 had written and compare it to Opus 5's code. Fable ran a dynamic workflow and the scores came back 15-20% higher for Opus 5's code in terms of quality, correctness and readability/conciseness. I did not tell Fable which Opus wrote which code, and I turned off memory as well to ensure there was no pollution from that angle.

      My only complaint is that Opus 5's prose is annoying as hell. I wrote a custom skill for it for concise debriefs and it has been working pretty well for me.

   8. fellowniusmonk · 1786047862 · focus · HN ↗

      I have some internal tests I use for areas where one particular solution/paradigm is dominant but worse.

      Opus 4.6 is the last model that's actually useful and can "adjust" its perspective to use he newer & better solution.

      Where Opus 4.8-5 has over fit training on worse/older but "dominant" solutions it refuses to adjust.

      Not only does this create an existential threat to adopting progress but it also means that if you have a code base that has rare but real world tradeoff Olthe newest versions of Opus are worse than useless but become a major dev timesink.

   9. Fordec · 1786048501 · focus · HN ↗

      Yeah, I've dropped back to 4.8 entirely for the remainder of this billing cycle. I'm going to be seriously looking into Qwen adoption and harness migration options over the course of August.
   10. cesarvarela · 1786050220 · focus · HN ↗

       It is infuriating to interact with, but it is also first in many blind test leaderboards on LLMArena
   11. CuriouslyC · 1786050223 · focus · HN ↗

       Ironically, Opus 5 is the most benchmaxxed model I've seen from Anthropic. It is legitimately smart in a lot of ways but it has communication issues, both in terms of how it communicates (all the autism of GPT class models, without the brevity) and how well it catches all the nuance of what you tell it.
   12. sunaookami · 1786050337 · focus · HN ↗

       Can not confirm, for me it's the complete opposite.

Unofficial Hacker News client; not affiliated with Y Combinator.