Particle.news

Moonshot Unveils Kimi K3, a 2.8 Trillion-Parameter Open-Source AI

Independent tests showing top-tier results with a one million‑token context indicate Chinese open‑source AI is narrowing the gap with U.S. leaders

Overview

  • Moonshot announced Kimi K3 on Friday, July 17, describing it as a 2.8 trillion‑parameter open‑weight model with a claimed 1 million‑token context window for much larger single‑session inputs.
  • Independent evaluators placed Kimi K3 at or near the top in several benchmarks, with Arena.ai, Vals AI and Artificial Analysis reporting strong results against high-end systems such as Anthropic’s Fable 5 and OpenAI’s GPT-5.5 on complex tasks and hardware-efficiency tests.
  • The launch triggered immediate market moves, with shares of several Chinese AI firms plunging on the Hong Kong exchange as investors reassessed competitive standings.
  • Practical use of a 2.8 trillion‑parameter model remains limited because running it requires very costly hardware and cloud resources that analysts say can cost hundreds of thousands of dollars, keeping full deployments to well‑funded firms or cloud providers.
  • Moonshot’s announcement, backed by investors including Alibaba and Tencent and tied to reported fundraising and Hong Kong IPO plans, highlights a broader trend of Chinese open‑source labs closing the performance gap and raising new questions for regulators after recent U.S. scrutiny of advanced models.