Kimi K3 Just Put Claude and OpenAI on Notice
By Miguel Torrez, founder of Genial Agency
You will know what Kimi K3 changed (an open source model near the top of the benchmark charts at about half the price of frontier models), why open source matters here, and how to check whether it fits your own work.
What you need
Access to Kimi K3 at kimi.com
A few real tasks from your own work to test it on
Optional: the model you use today (Claude or ChatGPT) to compare against
Step by step
Know what launched
Kimi K3 is an open source model. I call it the first open source model to rival, and on some results beat, Fable 5 and GPT 5.6 Sol, at about half the cost.
Check where it ranks
Look at the launch-day charts. On release day Kimi K3 was already in the top three in most of them, for both general use and coding use cases.
Note the UI benchmark result
Kimi K3 beat Fable 5 on UI benchmarks. The leading model for UI work is now an open source model from a Chinese company, not a closed source model from a frontier lab. If you build interfaces, test it there first.
Compare the pricing
Check the price before you switch. Kimi K3 costs five times more than the previous Kimi model, but it is still about half the cost of frontier models while delivering a similar level of quality.
Understand open vs closed source
Closed source: you pay the provider (OpenAI, Anthropic) and the models are private, so you cannot look inside them. Open source: the models are free and anyone can use them, but running them yourself takes very large computers.
See what it means for the big labs
Kimi K3 now sits in the top three next to Opus 4.8. My view: this puts more pressure on Anthropic, and Opus 5 has to beat Kimi K3 by a clear margin. For you as a user, more competition means better and cheaper models.
Test it on your own use cases
Open Kimi K3 and run it on the same tasks you give your current model. Test it as much as you can so you know whether it is the right model for your use cases.
Kimi K3https://www.kimi.com/
Watch out for
These are launch-day benchmark results. Rankings can change, so trust your own tests over the charts.
Kimi K3 costs five times more than the previous Kimi model. Do not plan your budget on the old model's pricing.
Open source does not mean easy to run yourself: today you still need very large computers to host a model like this.
