Kimi K3 is real: 2.8T params, 1M context, and a $3/$15 API. I ran the cost math and found the bigger catch is output verbosity.
Kimi K3 launched on July 16, and three claims immediately started traveling together:
"It uses all 2.8 trillion parameters on every token."
"The open weights are already available."
"At $3/$15 per million tokens, it is automatically cheaper per task."
Two are wrong. The third is incomplete.
Kimi's open model K3 nears GPT-5.6 Sol and Fable 5 while signaling the end of super cheap Chinese AI
China's Kimi K3 Matches America's Best AI At Half Price
Kimi K3 launches with 1 million context tokens, escalating the AI arms race that moves crypto markets
[AINews] Kimi K3 2.8T-A50B: the largest open model ever released; Opus 4.8-class at Sonnet 5 pricing
Kimi K3 launches as world's largest open-source AI model with 2.8 trillion parameters
Moonshot AI unveils Kimi K3, the largest open-weight AI model at 2.7 trillion parameters
China's Kimi K3 model stuns AI world with frontier-level results, and crypto AI projects are paying attention
What is Kimi K3, China’s first open AI model to reach 2.8 trillion parameters?
Kimi K3 shows open AI models have finally caught up with proprietary US-based rivals
Moonshot AI Releases Kimi K3: A 2.8 Trillion Parameter Open MoE Model With Kimi Delta Attention and 1M Context
Kimi is launching K3, a multimodal open-weight model with 2.8 trillion parameters and one million tokens of context. In the…
Moonshot AI's Kimi K3 is a 2.8-trillion-parameter open-source model with a 1M-token context window, native vision, and benchmark…
Moonshot AI's Kimi K3 is the largest open-weight model ever released - 2.8 trillion parameters, a million-token memory, benchmark…
Moonshot AI launches Kimi K3 with 1 million context tokens and Agent Swarm technology. Here's what it means for AI-adjacent…
Moonshot AI's Kimi 2.7 targets Claude Fable 5 with $0.95 per million input tokens and a trillion-parameter architecture for…
Moonshot AI releases Kimi K2.7-Code, a 1 trillion parameter open-source coding model with 30% fewer reasoning tokens and…
