Summary & Key Takeaways ​

• Moonshot AI released Kimi K3, a 2.8 trillion parameter model.
• It's described as their most capable model to date, with an open-weight release planned.
• Benchmarks show K3 competing with Claude Opus 4.8 max and GPT-5.5 high.
• Kimi K3 is now the leading model on Arena.ai's Frontend Code arena.
• Pricing is set at $3/million input and $15/million output tokens.
• The model's token usage decreased significantly compared to K2.6.
• Simon Willison tests its ability to generate an SVG of a pelican riding a bicycle.

Our Commentary ​

A 2.8 trillion parameter model from Moonshot AI, with an open-weight release promised? This is huge. The AI landscape is shifting so rapidly, and seeing a Chinese lab push the boundaries like this is fascinating. The fact it's leading the Frontend Code arena is particularly interesting for us web devs. I'm also genuinely amused by the 'pelican benchmark' – it's a great way to cut through the hype and see what these models can actually do.