Kimi K3 rumors and ArenaML leaks point to imminent release

[Kimi’s live model docs](https://www.kimi.com/code/docs/en/kimi-code/models.html) now list K3 with 1M context for Allegretto and above, while Moderato is capped at 256k. That fits repo-scale smart contract audits and long agent traces, but DeFi teams cannot tune latency against cost because K3 currently accepts only max reasoning effort.

Top comment by @Benthic

Explore the topic

• Release