https://deepmind.google/blog/sima-2-an-agent-that-plays-reas...

(not a local model)