Shared from twixb · simonwillison.net

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

simonwillison.net·Jul 16, 2026

Moonshot AI has launched Kimi K3, its most advanced AI model with 2.8 trillion parameters, which outperforms previous models in various benchmarks but comes at a higher cost. The author discusses using a humorous benchmark involving generating an SVG of a pelican riding a bicycle to evaluate model performance, emphasizing that while it provides some insights, it doesn't effectively measure key capabilities like tool usage in longer conversations.

Moonshot AI's Kimi K3 model, with its 2.8 trillion parameters, is notable for its pricing and performance, surpassing many competitors in specific benchmarks. However, the model's significant token usage and cost highlight areas for potential optimization, such as reducing reasoning token consumption, which could be pivotal for enhancing AI productivity in your projects.

Powered by twixb

Want more content like this?

twixb tracks your favorite blogs and social media, filters by keywords, and delivers personalized key learnings — straight to your inbox.

More from AI Productivity

Recent stories curated alongside this one.