Moonshot AI Launches Kimi K3: 2.8-Trillion-Parameter Open-Weight Model Tops Code Arena

Chinese AI lab Moonshot AI just dropped a bombshell: Kimi K3, a 2.8-trillion-parameter model that is now the largest open-weight model in existence — and it beat Anthropic's flagship on code.

Three Key Takeaways

#1 on the Frontend Code Arena. K3 debuted at the top of Arena.ai's Frontend Code blind evaluation with an Elo of 1,679, overtaking Claude Fable 5 and claiming first place in six of seven frontend domains. For context, the previous Kimi K2.6 ranked 18th — a 17-place leap in a single generation.

Largest open-weight model ever. K3 is a 2.8-trillion-parameter Mixture-of-Experts (MoE) model with a 1-million-token context window, native multimodal input, and max reasoning at launch. Moonshot calls it "the world's first open 3T-class system." Full weights are due by July 27.

Overall gap remains. Moonshot itself acknowledges that K3 still trails Claude Fable 5 and GPT-5.6 Sol on comprehensive benchmarks, but it has surpassed Claude Opus 4.8 and GPT-5.5 across coding and agentic evaluations.

WangDou's Take

Moonshot played this smart: instead of chasing the overall leaderboard against closed-source giants, they picked coding — the one arena where developers can verify claims with their own eyes. The 2.8 trillion parameter count sounds terrifying, but MoE means only a fraction fires during inference, keeping costs manageable. The real number to watch is that 17-place jump — from K2.6 at #18 to K3 at #1. That is not incremental improvement; that is a discontinuity. If the full weights actually ship on July 27, every company that wants an in-house AI coding assistant without paying Anthropic or OpenAI rent is going to run K3 through its paces. The open-weight arsenal just got a lot heavier.

Source: Tom's Hardware, Tech Startups