You don't need anybody to permit you access.
You can, in all seriousness, thank Meta and the Chinese for this.
You don't need anybody to permit you access.
You can, in all seriousness, thank Meta and the Chinese for this.
What open models are "a few months behind" Claude Code Opus, Google Nano Banana Pro, or ByteDance Seedance 2.0?
Another big problem with the open models is that the ecosystem targets consumer GPUs rather than running in the cloud. The future is a thin client world with stuff at the edge being generations behind in performance.
Kimi K2.5 is quite literally just a few months behind, at worst. In many cases it already outperforms Opus 4.5 (and reliably outperforms Sonnet), and it's not really that far from Opus 4.6.
I'm less familiar with the video/image side of things.
> Another big problem with the open models is that the ecosystem targets consumer GPUs
No, this is something of a misconception. It's quite difficult to run a model like K2.5 truly locally. But any nation-state or sufficiently large corporation can get its hands on HPC clusters and run it freely, without needing permission from anybody.