AI Disruption

AI Disruption

Kimi K2 Thinking Launched, Outsmarting GPT-5

Kimi K2 Thinking: open-source 1T-MoE agent model beats GPT-5, MIT weights live on kimi.com

Meng Li's avatar
Meng Li
Nov 07, 2025
∙ Paid

“AI Disruption” Publication 8100 Subscriptions 20% Discount Offer Link.


NEW Kimi K2 Thinking - Think Harder - YouTube

Kimi K2 Thinking is now released and open-sourced!

Featuring a “model as agent” approach, it’s not only Kimi’s “most capable open-source thinking model to date,” but also masters the ability to think while using tools simultaneously—

Without human intervention, it can execute 200-300 consecutive tool calls.

As one of this year’s most anticipated open-source model series, the Thinking version of Kimi K2 became a hot topic upon launch, once again narrowing the gap between open-source and closed-source models.

Key Technical Details:

  • 1TB parameters, 32B activated parameters, using INT4 instead of FP8

  • 256K context window

  • More experts, fewer heads, more thinking

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Meng Li
Privacy ∙ Terms ∙ Collection notice
Start your SubstackGet the app
Substack is the home for great culture