Overshot — find the TikToks breaking out, and what they're selling

Kimi K3: acts - it is a large AI model that processes information and is being benchmarked and discussed. Removing it would change the video's topic. Moonshot AI: acts - it is the company that released Kimi K3. Removing it would change the video's topic. DeepSeek V4 Pro: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Fable 5: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. GPT-5.6 Sol: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Opus 4.8: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Opus 4.7: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Opus 4.5: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Opus 4.0: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Sonnet 2.0: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Sonnet 1.5: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Sonnet 1.0: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Pro: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. GLM-5.1: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Claude Sonnet 4.0: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Owen-3.7 Max: acts - it is an AI model that is being compared to Kimi K3. Removing it would change the video's topic. Hugging Face: acts - it is a platform where AI models are released. Removing it would change the video's topic. Gameboy Advance: appears - it is shown as a visual example of a retro gaming device. Removing it would not change the video's topic. Gamecade: appears - it is shown as a visual example of a retro gaming device. Removing it would not change the video's topic. Kimi K3: acts - it is an AI model that is being benchmarked and discussed. Removing it would change the video's topic. Frontend Code Arena: acts - it is a benchmark for AI models. Removing it would change the video's topic. SWE Marathon: acts - it is a benchmark for AI models. Removing it would change the video's topic. Program Bench: acts - it is a benchmark for AI models. Removing it would change the video's topic. Cloudflare: appears - it is the title of the linked page, but not directly relevant to the video's content. Removing it would not change the video's topic.

Transcript

The largest AI model in open-source history just released. It’s another massive frontier model from Moonshot AI in Beijing, China. Kimi K3 has 2.8 trillion parameters and comes with an entire suite of technical and logical upgrades. It leads on multiple benchmarks and directly challenges Claude Fable 5 and GPT 5.6 Sol. The release comes with a looming threat for competing frontier labs - the raw model weights for Kimi K3 are scheduled to be released on Hugging Face July 27th! #breakingnews #chin I've already told you about Kimmy. This new iteration comes with some surprises. K3 comes with 2.8 trillion parameters. That's nearly double DeepSeek V4 Pro and it eclipses every other open-weight model on the planet. Once again, it uses MOE design. And if you follow me, you know that mixture of experts means only a slice of those parameters fire when you ask a question. You're getting frontier size with a frontier cost. And yeah, if you've got the Kimi app on your phone, you can start using that model for free. Yes, the free version is rate limited and yes, some people have security concerns, which I respect. But you should find a way to try this out. It's seriously that good. We've seen AI models build games like this before. Have you seen 3D emulators where you can play a game on a nostalgic handset that's also generated by the AI model? Maybe you have. But have you seen a cinematic generated video where an AI model builds its own chipset right in front of you? "By Kimi K3, for Kimi K3." That's pretty cool. It's got a million token context window. And we know that's impressive, but we also know it's not exactly groundbreaking. Rather than having to fake performance with multiple agents, this thing can process your entire codebase in one swoop. (Depending upon) that makes decoding up to 6.3 times faster at full context. It's speedy. Now let's talk proceeds. 91.2 on BrowseComp. The highest ever recorded by any model, period. Number one on Frontend Code Arena, Kimi K3, ranked number one. SWE Marathon, Program Bench. It's right on the heels of Fable 5 and GPT-5.6 Sol. And like I said, it's live right now. But here's the real kicker. Word is that on July 27th, Moonshot AI is dropping the full raw model weights open source on Hugging Face. For years, the rule has been open source trails proprietary tech by six months. That's over now. Weeks ago, we were forbidden from accessing this tier of technology. In a couple weeks, it's open. We can run it locally. All we need is a setup that is laughably impressive. It's over now.

The largest AI model in open-source history just released.

Selling · SaaS › AI Tool

Why it worked

Summary

Structure

Product placement

Original caption

More from @gigaqian

On-screen text