Introducing Kimi K3: Open Frontier Intelligence
🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal
🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts
🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost
🔹 Built for long-horizon agentic coding and self-evolving workflows
Kimi K3 is now live on on https://t.co/zrk6zZxZUo, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.
🔗 API: https://t.co/XCrgjXAqMw
🔗 Tech blog: https://t.co/YTfiMSNM1f
Human potential is infinite. As long as we keep engaging in relentless questioning and keep asking an interesting set of questions, the sky is the limit in terms of what we can actually learn.
owning the infrastructure stack pays off. Better latency, more control, and less dependency on third-party providers, great example of engineering at scale.
We’ve built our own sandbox lifecycle management platform SPACE to serve all production traffic of Perplexity Computer.
When we rolled out Computer in February, we ran it on one of the sandbox providers. Today, we serve 100% of the traffic on SPACE with 5x faster tail latency.
The ability to be curious, to learn across a lot of disciplines and to have a strong foundation of wanting to have impact, regardless of the area that you're working in, I think that is an underrated quality.
Faster agentic coding on CPUs is a bigger deal than many realize. if these gains hold across broader workloads, this could meaningfully reduce AI infrastructure costs and improve developer productivity. Looking forward to the detailed Vera CPU benchmarks.
@AravSrinivas The third path might be improving model efficiency ,better architectures, quantization, sparsity, and distillation could reduce power demand before we move compute off Earth.
I think this is one of the most underrated predictions in AI. The future may not be one giant model,but small, capable models orchestrating the best tools available.
Humans are pretty good at tool use. Especially using tools like frontier models that are far more power hungry and intelligent than humans in specific dimensions. This suggests that local models will be able to control and use frontier models effectively and will become the default entry point to operate most tasks by running at minimal power and costs.
ChatGPT Work brings agents to consumer scale.
It’s both a step up in usability (you can just do things from your phone, no laptop required) and access, for both personal and professional life.
Very excited to see how people put it to use and how it can empower them!