Google has abandoned their Ultra series and it seems OpenAI has no plans to release a ~2T model like GPT-4 was. I think it's safe to say Opus has the same fate.

And yes, I know Dario said "as far as we know, the plan is still to have a Claude 3.5 Opus", but this is far from a strong claim and IMO a deflection from the fact that these large models just aren't much better than 100b-500b ones. The reality is scaling model size and pretraining isn't working anymore and he doesn't want to admit it because it makes Anthropic look bad.

Dunno, the fact that they've gone with a second release of sonnet 3.5 is definitely curious

Because if Opus 3.5 was good you'd expect them to release that first with a flashy blog post, and then go with the new sonnet 3.5 as a follow up.

The whole "sonnet 3.5 new" nomenclature fiasco screams plan B

The only way this would make sense to release Opus 3.5 between now and the end of the year would be if it was truly crazy good and they just needed a bit more time to get it right. But all else being equal, that doesn't seem super likely.

Another possibility is that anthropic doesn't care much about their 3.x models any more and fully switched to 4.x work a few months ago. These releases might just be about getting remaining capabilities of this tier out the door without too much fanfare, apart from the computer use stuff.

If it becomes available before 2025, but then is discontinued how would you resolve this?

@JoeBoyle Yes