The K2 architecture? 1T total parameters.
32B active MoE.
Optimized for reasoning, not just response speed. This is what “thinking AI” looks like.
TECHNOLOGY
-
K2 architecture with 1T parameters and 32B active MoE for reasoning
By
–
-
Open-source Kimi model outperforms GPT-5 using 1/1000 compute
By
–
Kimi just released the best open-source thinking model last week. It topped benchmarks like HLE and BrowseComp, even beating big closed-source models like Grok-4 and GPT-5. And get this: they pulled it off using just 1/1000 of the compute those closed models use. Absolutely
-
xAI drops Grok-4-Fast: 2M-token reasoning model, faster, cheaper
By
–
xAI just dropped a monster: Grok-4-Fast Its a 2M-token reasoning model that’s faster than anything in its class and cheaper than lunch. $0.20 in / $0.50 out per million tokens. That’s 6× cheaper than Grok-4-0709 with the same brain. The wild part? It’s not just speed. It’s
-
AI-based climate emulation with Chris Bretherton livestream
By
–
Chris Bretherton talking about AI based climate emulation. Join our livestream https://
aiscienceconference.caltech.edu -
Heart and Vascular Care Wins Platinum Award for Healthcare Website
By
–
Congratulations to Heart and Vascular Care at @BaptistHealthSF for winning a Platinum Award for Best Overall Internet Site at the #eHealthcareAwards! https://
hubs.ly/Q03SCLrF0 -
US AI Ecosystem Advantage Over Single Technology
By
–
When I asked Dr. Craig Martell, CTO of Lockheed Martin, how he sees 🇺🇸 in the context of the global technology race, he told me that the real asymmetry the US holds is not any ‘one’ technology.
— Nina Schick (@NinaDSchick) 11 novembre 2025
Rather it is the U.S. ecosystem itself: capital, academia, private enterprise,… pic.twitter.com/regrwYzmrZWhen I asked Dr. Craig Martell, CTO of Lockheed Martin, how he sees in the context of the global technology race, he told me that the real asymmetry the US holds is not any ‘one’ technology. Rather it is the U.S. ecosystem itself: capital, academia, private enterprise,
-
Templates on Sora, Veo, Aurora auto-generate product variants with hooks and CTAs
By
–
Each template runs on Sora 2, Veo 3.1, or Aurora depending on what you need. Cinematic product shots → Sora/Veo
UGC testimonials → Aurora You change your product once and it auto-generates variants with different hooks, CTAs, and platform ratios. -
AI Transforming Telecommunications and Next-Generation Wireless Networks
By
–
Missed #NVIDIAGTC Washington D.C? Our telecommunications sessions are now available on-demand. Watch now to explore how AI is transforming the industry and every layer of next-gen wireless. https://
nvda.ws/47KsoOw -
Grok 4 Fast has 2M token context window, 5x ChatGPT, 2x Gemini
By
–
Context window comparison: • ChatGPT 5: 400,000 tokens
• Gemini 2.5 Pro: 1,000,000 tokens
• Grok 4 Fast: 2,000,000 tokens that’s 5x bigger than ChatGPT. 2x bigger than Gemini. load your entire codebase as one prompt. every doc. every conversation. every data point. no -
4K Video Export Zoomed Still Frame Production Technique
By
–
Step One
— Runway (@runwayml) 10 novembre 2025
Import your 4K footage to a timeline and export a zoomed in still frame.
(2/6) pic.twitter.com/VkzmjO23TKStep One Import your 4K footage to a timeline and export a zoomed in still frame. (2/6)