RC RANDOM CHAOS

Moonshot's Kimi K3 hits 2.8T parameters — and Simon Willison's pelican test bows out

· via Hacker News

Original source

Kimi K3, and what we can still learn from the pelican benchmark

Hacker News →

Chinese lab Moonshot AI unveiled Kimi K3, a 2.8-trillion-parameter model it markets as the first ‘open 3T-class’ system, with open weights promised by July 27, 2026. Self-reported benchmarks put it ahead of Claude Opus 4.8 and GPT-5.5 but behind Claude Fable 5 and GPT-5.6 Sol, and third-party evaluations from Artificial Analysis show strong long-horizon results plus a 21% drop in output-token usage versus its predecessor. The bigger story is pricing: at $3 per million input and $15 per million output tokens, K3 is the most expensive model a Chinese lab has shipped, roughly triple the cost of Kimi K2.6 and on par with Anthropic’s Sonnet tier.

Read the full article

Continue reading at Hacker News →

This is an AI-generated summary. Read the original for the full story.