Moonshot Drops Kimi K3, a 2.8-Trillion-Parameter Open-Weight Frontier Model — and Puts a Date on the Weights
The Chinese lab released a natively multimodal model with a 1M-token context window and promised open weights by July 27, immediately topping a frontend coding leaderboard and reopening the debate over whether US lab economics can survive commoditized frontier capability.
Moonshot AI launched Kimi K3 on Thursday, and the specifications alone were enough to reset the week's conversation. In its announcement, @Kimi_Moonshot described the model as "Open Frontier Intelligence" — 2.8 trillion parameters, a 1-million-token context window, native multimodality, and a commitment to release open weights by July 27, 2026. That last detail is the one that matters most. A frontier-scale model with published weights is a fundamentally different economic object than a frontier-scale model behind an API.
The performance claims did not stay theoretical for long. According to @dons_korea, K3 has already overtaken Claude Fable 5 to take the top spot on the Frontend Code Arena leaderboard. If that result holds under independent testing, it means an openly released Chinese model is now leading a competitive coding benchmark that Western labs have treated as a core moat. Coding is the workload that pays the bills for most commercial LLM providers. A free-to-run model that wins there is a direct threat to margins.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.