Moonshot AI’s open-weight model Kimi K3 has become a talking point in Silicon Valley after its release on July 16, with Business Insider highlighting industry reactions. The model has 2.8 trillion parameters and is billed as capable of processing hundreds of pages at once. It also tops Arena’s front-end coding leaderboard and is priced below comparable offerings. Some executives and academics praised its performance, while others urged caution about relying on benchmarks.

Moves to develop AI models optimized for robots are accelerating, with two camps competing for technological leadership: vision-language-action models derived from large language models, and world models trained mainly on video to predict what will happen in real environments. The Information reported growing Silicon Valley interest in world models, though critics say they still make frequent errors. Some say VLA remains too slow and unreliable for broad production use.