The biggest open sourced AI model just dropped

@ai.samaritan

12.4K followers · usually 3.8K views a post

Views

69.9K

Likes

3.2K

Comments

51

Shares

347

SoundTikTok sound · Vivaldi-Four Season-Summer-3

Promoted product

Kimi AI

Try Kimi K3 to build playable multiplayer and 3D games, create consulting grade slides, and run parallel tasks with Swarm and Goal to get more work done.

62 videos51 creators

Why it worked

The video effectively highlights a significant advancement in open-source AI by showcasing Kimi K3's top ranking against established closed models. It leverages the 'open vs. closed' narrative and provides concrete performance data, making the information compelling for AI enthusiasts and developers.

Summary

A new open-sourced AI model called Kimi K3 has been released by Moonshot AI, achieving the top rank in the Frontend Code Arena and surpassing closed models. The model's development involved self-optimization, and it will be fully available to the public on July 27th.

Structure

1Introduction of Kimi K3 as the biggest open-sourced AI model
2Kimi K3 ranks #1 in the Frontend Code Arena, surpassing competitors
3Discussion of Kimi K3's development and self-optimization capabilities

Original caption

The gap between open and closed AI just got uncomfortably small and this time it's not from DeepSeek #aiupdates #ainews

More from @ai.samaritan

Product placement

Kimi K3: acts - an open-sourced AI model that ranks #1 in the Frontend Code Arena. Moonshot AI: acts - the company that developed Kimi K3. Arena.ai: acts - a platform that ranks AI models. Claude Fable 5: acts - a closed AI model that Kimi K3 surpasses. GPT-5.6 Sol: acts - an AI model mentioned in benchmarks. GPT-4.5: acts - an AI model mentioned in benchmarks. GPT-3.5: acts - an AI model mentioned in benchmarks. GLM-5.2: acts - an AI model mentioned in benchmarks. Muse Spark 1.1: acts - an AI model mentioned in benchmarks. Seed-2.1 Pro: acts - an AI model mentioned in benchmarks. GLM-5.1: acts - an AI model mentioned in benchmarks. Qwen-1.5-72B: acts - an AI model mentioned in benchmarks. Kimi K2.6: acts - a previous version of Kimi K3. Practicaly.ai: acts - the creator's website, mentioned as a place to get AI training.

Call to action

Visit the Kimi API Platform and select k3st-k3. Visit the pricing page and select "Get Kimi Enterprise" to subscribe for your team.

On-screen text

The biggest open sourced AI model just dropped THE BIGGEST OPEN MODEL IN HISTORY JUST DROPPED AND IN 11 DAYS CAN DOWNLOAD IT. THIS IS KIMI MOONSHOT AI AND IT'S THE FIRST OPEN MODEL TO CROSS 2.8 TRILLION PARAMETERS. ALONE DOESN'T MEAN MUCH SO HERE'S WHAT ACTUALLY MATTERS. THE NUMBER ONE FRONTEND CODE ARENA BEATING EVERY CLOSED MODEL INCLUDING CLAUDE'S NEWEST FLAGSHIP THAT THAT'S A PLACE TO FIRST IN A SINGLE GENERATION. AND TO MOON WITH HONEST IN THEIR OWN BLOG. OVERALL IT STILL TRAILS CLAUDE FABLE PT 5.6 SOL BUT THAT'S THE BETWEEN THE BEST OPEN MODEL AND THE BEST MODEL A YEAR. IT'S OF BENCHMARKS AND WEEKS. AND HERE'S THE DETAIL WORTH THINKING ABOUT. DURING DEVELOPMENT, AN EARLY VERSION DOING THE MAJORITY OWN GPU KERNEL OPTIMIZATION. THE MODEL WAS HELPING BUILD ITSELF. IT'S LIVE TODAY IN AP FULL WEIGHTS GO IN PUBLIC JULY 27TH MEANING ANYONE ANY STARTUP, ANY LAB, ANY COUNTRY GETS FRONTIER ADJACENT INTELLIGENCE WITH NO RESTRICTIONS.

Transcript

The biggest open sourced AI model in history just dropped and in 11 days anyone on earth can download it. This is Kimi K3 out of Moonshot AI and it's the first open model to cross 2.8 trillion parameters. Now, parameter count alone doesn't mean much, so here's what actually matters. It just took the number one spot in the Frontend Code Arena beating every closed model including Claude's newest flagship. That that's a jump from 18th place to first in a single generation. And to Moonshot's credit, they're honest in their own blog. Overall, it still trails Claude Fable 5.6 Sol, but that's the story. The distance between the best open model and the best model a year ago. It's now a matter of benchmarks and weeks. And here's the detail worth thinking about. During development, an early version of Kimi K3 was doing the majority of the team's own GPU kernel optimization. The model was helping build itself. It's live today in API, full weights go public July 27th, meaning anyone, any startup, any lab, any country gets frontier adjacent intelligence with no restrictions.