This Chinese AI just beat ChatGPT

@maverickgpt

587.1K followers · usually 7.7K views a post

Views

34.5M

Likes

264.1K

Comments

4.4K

Shares

21.1K

SoundTikTok sound · original sound - maverickgpt

Promoted product

K2

Kimi K2.5 is an open source multimodal model by Moonshot AI, combining visual coding, AI agents, and agent swarm capabilities for real work.

3 videos3 creators

Why it worked

The video leverages a 'breaking news' hook about a high-stakes AI race, combined with a clear, impressive 'before and after' demonstration of the product's utility, making the promotion feel like valuable industry analysis.

Summary

The creator discusses a new Chinese AI model called Kimi K2 Thinking, highlighting its superior performance on benchmarks and lower training costs compared to ChatGPT. He demonstrates the tool's capability to convert a 100-page PDF into a visual PowerPoint presentation using its 'slides mode'.

Structure

1Hook with benchmark chart
2Introduction of Kimi K2 Thinking
3Comparison of training costs and efficiency
4

Original caption

China just overtook US in the AI race 🤯 Moonshot AI dropped a model called Kimi K2 Thinking and it beat ChatGPT, Gemini, and Grok on all the major benchmarks! What do you think about this? #chatgpt #ai #kimik2 #nanobanana #llm

More from @maverickgpt

Product placement

The video features the Kimi K2 Thinking interface throughout the demo, showing its capabilities and specific features like 'slides mode' powered by Nano Banana Pro.

Call to action

let me know what you think

On-screen text

This Chinese AI just beat ChatGPT Humanity's Last Exam (Text-only) w/ tools Kimi K2 Thinking 'Another DeepSeek moment': Chinese model Kimi K2 stokes excitement KIMI Open source Closed source Let's Run Local AI Kimi K2 Thinking - Chart Topping 1 TRILLION Parameter Span Model The new Kimi AI model cost $4.6 million to train, according to a source familiar with the matter. The initial training costs for OpenAI's foundational large language models ranged from millions of dollars for earlier versions to over $100 million for more advanced models like GPT-4. KIMI High-detail breakd... LLM MECHANISMS UNWOUND THE FIRST SUB-ASSEMBLY: TOKEN LATTICE & POSITIONAL ROPES MULTIPLE HEADS: PARALLEL LATHES FOR THOUGHT PROCESSING Kimi K2 Thinking GPT-5 Claude Bonnet 4.5 (Thinking)

Transcript

This Chinese AI just beat ChatGPT. It just dropped and it's beating almost every benchmark and no one's talking about it. It's called Kimi K2 Thinking, and experts are calling this another DeepSeek moment. This is the interface right here and it looks like ChatGPT but it's actually 10 times smarter. It's completely open source with over a trillion parameters and people are already running this thing straight from their Mac Studios. To put that in a perspective, almost every AI model runs on massive data centers. And it only took $4.6 million to train, which is a lot less than ChatGPT's $100 million. To test it out, I turned on slides mode which uses Nano Banana Pro, and then uploaded a 100 page PDF and told it to turn this into a visual PowerPoint. Most AIs would have just broke down here, but check this out. This would have taken any designer weeks to make, but Kimi did it in less than two minutes. People in the AI space are already freaking out about this, so let me know what you think.