upvote
> I find it weird there's not more discussion here on HN on how the most used model now has clearly degraded in quality and it seems we've hit a peak and are on a downslope

It's not weird, because it's an anecdote, not an accepted fact.

Personally I've not been too happy with Opus 5, but I've had similar experiences with other models previously, feeling like they didn't quite fit with my working style.

So nothing indicates we've hit a peak.

reply
I'll say everything indicates we've hit or are near peak for the masses at least (unless you start paying 50x more) but to each his own.

4.6 was best for us and right now yeah OpenAI and others are edging forward, but slower while prices are increasing industry wide as much as 20x, time to completion is increasing wildly and i'm sure they'll do the same over at OpenAI as their compute constraints also start to take a toll ie degrade performance.

In my view 4.6 era was way faster and with less weirdness so we've gone downwards at least in my company, 4.7 was ridiculous, then 4.8 was almost 4.6 level, 5 is even worse than 4.7 - so it's not a bit up and down its down then a little up then further down.

And all of this is against a backdrop of zero ROI in this sector - so it makes sense we've hit a peak and we're now seeing the subsidisation phase begin to falter, will there be better models certainly but only for short amounts before they get quantised (or whatever is happening behind the scenes), and with diminishing returns over huge prices increases and slower responses.

reply
Literally nothing indicates we've hit a peak, but I guess I'm discussing this with someone who thinks every iteration since Opus 4.6 had zero ROI so there's probably not much common ground here.
reply
> nothing indicates we've hit a peak

Opus 5 is arguably a regression but GPT 5.6 is pretty strong evidence that we haven't hit a peak. I think I actually prefer Sol to Fable at this point.

reply
I have a chunky bit of functionality in my hobby app using babylon.js to render 3D worlds using things like portals and LoD rendering to manage the visual load. I built it out with a combo of Fable and Opus 5.

I too got fed up with the prose of Opus in particular, and tried going back. Unfortunately, the previous models were less able to hack it. The prose was better but progress was worse.

It wasn't just conversation and comments. Some of the function names were wild. Like it instead of something like "isSolidWall(x)" it would write something like "weightyNotEphemeral(x)" or something - that's not quite it, but it really did embed overwrought antithesis into the identifier instead of a straightforward positive predicate.

reply
Yeah. I set my default back to 4.6 and only use 5 for code review etc. Also save s alot of tokens...
reply
We can use OpenRouter pricing to get an idea about what competitive inference pricing is like without R&D or other costs, and indeed we'd be screwed if we had to pay those rates. We'd go from 100-200 USD to 2000-4000 USD/m.
reply
Not sure what you're talking about here. Kimi K3 is frontier scale and sells competitively at $2.80 input, $14 output per 1M tokens.

edit: oh you mean month? Sure, but then it fully depends on your usecase. I agree that subscriptions are heavily subsidized though.

reply
no if prices were that high, millions of developers would switch to hand writing, open source LLM help and outsourcing at those levels, AI companies know this, you should too.
reply
It is bizarre. There has been no statement, no mention of even hearing concern about Opus 5.

I presume something is forthcoming, but it may be they don’t want to come empty handed—-5.1 is intended to “fix the glitch.”

reply
yeah I feel the same way with Opus 5 too. If I ask it to do something, it would go ahead and rewrite unrelated things and then in a less performant version of it.

I’ve instead moved to GLM, at least it has the courtesy to ask some steps of the way what I wanted exactly and only work on what I asked.

reply
>it seems we've hit a peak and are on a downslope

Sol and Fable are great; we haven't hit a peak, Anthropic just tried to pull a fast one on its customers with Opus 5.0.

reply
Opus 5 has a habit of taking what I asked for, doing something tangentially related to it, and then lying to me and saying it did exactly what I asked.
reply
It even adds comments to say you asked for [thing] then leaves snarky comments about when you correct it.

At that point I decided it's just not worth the babysitting that's required, and you are better off working entirely with other models.

If the harness itself was open source then maybe we'd be able to wrap it up in a reasonable layer of sanity.

reply
[flagged]
reply