Moonshot AI's Kimi K3 Arrives with 2.8 Trillion Parameters, Targets Frontier Model Rivals

Moonshot AI has officially previewed its Kimi K3 model, a large language model boasting 2.8 trillion parameters and a 1 million token context window, according to developersdigest.tech. The model employs a new Delta Attention architecture and is positioned to compete directly with top-tier models from Anthropic and OpenAI, including Claude Opus 4.8 and GPT-5.6. TechCrunch, citing the Financial Times, reported that Kimi K3 is expected to be the largest open AI model from China, with a parameter count between 2 trillion and 3 trillion. The Financial Times itself stated that Kimi K3 “exceed performance of Claude Opus” in benchmarks, though exact comparisons remain under wraps. "K3 rivaling Fable 5 in video generation detail and animation smoothness." Pandaily noted that users testing the model on the Arena benchmark have observed Kimi K3 rivaling Fable 5 in video generation detail and animation smoothness. This suggests Moonshot is pushing beyond text-only capabilities into multimodal AI. The company teased the model with promotional videos on Bilibili and X, generating buzz across social platforms, according to toolify.ai and digg.com. AI commentators have begun debating how Kimi K3’s performance and pricing could reshape the competitive landscape between Chinese and American AI firms. Crypto Briefing argued that the launch could “intensify US-China tech rivalry” and boost decentralized compute networks, as developers may seek low-cost alternatives to Western cloud providers. With Moonshot AI’s aggressive pricing for Kimi K3, the frontier model race is entering a new phase where parameter count and context window size are no longer the only battlegrounds.

7 sources•17 JulRead the Baker's Take→