> Plateauing of model capability improvement

On what are you possibly basing this? The fact that there hasn't been a new model release in 6 weeks?

Have the models become significantly more capable in the last 6 months? I'm not convinced we're still seeing massive improvements as we saw at the start. The output certainly isn't getting better, the software these companies produce is actually pretty buggy.

▲

origini an hour ago | parent | next [-]

For me personally? yes absoluttly.

Opus 5 vs. 5.5 is a clear stepup.

Problem is, you need to do a diverse set of experiments to recognize it. Blender support got relevant better,t he amount of corrections got less etc.

	▲	grey-area an hour ago \| parent [-]
		It does very much vary with the domain though. I haven't personally noticed much improvement, they're still useful but unreliable, in my view the idea of agents, harnesses etc has been a distraction and functions to cover up the lack of progress in the underlying models. But everyone has a different view of this, it'll be interesting to see how it plays out.

▲

Aeolos 2 hours ago | parent | prev [-]

Yes, without a question. Fable 5 and Opus 5.5 are a step-change above what we had before. On the open-weights side, GLM 5.3, DeepSeek 4.1 v1017, Qwen3.8-Flash-Next are all significantly more capable in basically any task you can give them.

You can now have a model work on a task non-stop for a day or more. This was unthinkable 12 months ago, and unstable 6 months ago.

If you are searching for evidence of capability, just look at the amount of projects, blog posts and videos on game remixes, hardware reverse engineering, new compilers and runtimes that has started coming out in the last 6 months. Slop? Sure, but this kind of slop didn't exist at this level 6 or 12 months ago.

	▲	yladiz 8 minutes ago \| parent \| next [-]
		Are you certain this is a model improvement and not improvements to agents?
	▲	grey-area 2 hours ago \| parent \| prev \| next [-]
		And yet we have to contrast that with reports like this of model capability falling recently: https://news.ycombinator.com/item?id=50422069 It's really very hard to pin down progress in this domain, companies are definitely targeting benchmarks. One bit of evidence I would look at is the maths results this last year which have been impressive, that was a huge expensive brute force effort, but it is still good progress.
	▲	pixelstack an hour ago \| parent \| prev [-]
		i am wondering in what potential world would average user or even business user require for model to work on a task for more than a couple of hours? Is assumption that models on average are used to build nextt SpaceX or something?