LithosAI says Kimi code model runs at 1,000 tokens per second
LithosAI presents Lithos Engine as a way to run Kimi K2.7 Code at 1,000 . Kimi K2.7 Code is a with 1 trillion , running on 8 B200 machines.
It is for use with Claude Code, Codex, or a developer’s own coding setup. LithosAI compares this with 174 to 291 from major providers, making its setup 3.4 to 5.7 times faster.
The company says the speed gain does not use approximation and keeps the model’s native precision and full quality. LithosAI was founded by Carnegie Mellon computer science professors Dimitrios Skarlatos and Zhihao Jia.
Key points
- LithosAI says Kimi K2.7 Code can run at 1,000 .
- The speed claim is 3.4 to 5.7 times faster than major providers.
- It is meant to work with Claude Code, Codex, or a developer’s own coding setup.
- The model has 1 trillion and runs on 8 B200 machines.
- LithosAI says it keeps full model quality without approximation.