>>106229584
I am in fact here and monitoring the Mistral situation but I reiterate that there's no info about upcoming DeepSeek releases. My prediction for something by end of August (27th specifically) stands. In principle V4 can be as late as October. They are in complete radio silence, except for the recent ACL presentation by Jingyang Yuan which seems to confirm a 1M context NSA model. I expect it to be at least as big as V3, even more optimized for high-bandwidth systems and useless for local.

Argentinians are annoying but less annoying than people who ask me about DeepSeek as if I'm a berryboi leaking alpha from secretive labs.