Dev.to
7/27/2026
Moonshot AI ships Kimi K3: 2.8T open-weight model tops coding benchmarks, with caveats on reasoning-token costs
Original: Kimi K3 Is the Biggest Open-Weight Model Ever Shipped. Here's What Actually Matters.
Short summary
Moonshot AI released Kimi K3, a 2.8T-parameter open-weight MoE model that tops independent coding benchmarks and matches frontier closed models on quality while undercutting them on price. However, Simon Willison's testing reveals significant reasoning-token overhead that can inflate real-world costs well beyond sticker prices for agentic workloads. The release signals that the gap between open-weight and closed models has effectively closed, with full weights available on Hugging Face and an OpenAI-compatible API for easy integration.
- •Kimi K3 is a 2.8T open-weight MoE model from Moonshot AI, #1 on Frontend Code Arena and best-ever GPQA Diamond score for open weights
- •Flat pricing at $3/M input and $15/M output across 1M context, but reasoning-token overhead can dramatically increase real costs
- •OpenAI-compatible API and Hugging Face weights make adoption trivial; pilot for agentic coding but test actual token spend before committing
Generated with AI, which can make mistakes.
Is this a good recommendation for you?
