I wonder if we are seeing them maxing out llm model capabilities now...
6.1 made a lot of assumptions and would "just do it." For many professionals that was a bad thing. For mnay users it was a good thing.
6.2 on the other hand does what it's told, and uses less flowery language.
I think they're focusing more on the enterprise.
I haven't tried the new Fable yet.
But I wasn't talking about subjective experience I was talking about benchmarks like artificial analysis for coding agents, where it gets slightly lower score than 6.1 (frontier code 2.0 it gets a 2 point drop).
It was 100% different though, and it was a bit jarring. It was kind of like having one of your coworkers replaced with a similar, but different, pod-person. I'm getting used to that feeling though.
The one thing I think we can all agree is certain is that it will be different in a few months.
I kind of wish they had LTS versions of some models, so you could build processes and tooling around them with certainty that they'd be exactly the same in 5 years.