The video discusses a significant development in the field of open-source AI models, specifically highlighting a new Chinese model with 2.8 trillion parameters and a million-token context window. This model has achieved a leading position on the "arena front end code leader board," surpassing well-known models like Fable and even GPT-4 (indicated by "GPT 5.6" which is likely a typo for GPT-4 or a similar high-tier model). The speaker expresses astonishment that this "frontier class model" is being released as open-source by the company Moonshot, emphasizing that it is essentially being given away for free. The speaker conveys excitement and eagerness to utilize this powerful new resource.

Matthew Berman

Chinese open source may have actually caught up. This is a 2.8 trillion parameter open weights model, million token context window, and is in first place on the arena front end code leader board, ahead of Fable, ahead of GPT 5.6 Soul.

This is truly a frontier class model that the company Moonshot is just giving away. It's amazing. I cannot wait to use it.

Kimi K3 just beat FABLE.

Chinese open source may have actually
caught up. This is a 2.8 trillion
parameter open weights model, million
token context window, and is in first
place on the arena front end code leader
board, ahead of Fable, ahead of GPT 5.6
Soul. This is truly a frontier class
model that the company Moonshot is just
giving away. It's amazing. I cannot wait
to use it.

Get Our Mobile App

Kimi K3 just beat FABLE.

Transcription