Kimi 3: China's Open-Weight Model Matching Anthropic Opus 4.8

Chinese lab Moonshot AI is preparing to release Kimi 3, a language model with 2–3 trillion parameters. According to the Financial Times, the model is expected to match or surpass Anthropic Opus 4.8, the current flagship among commercial closed systems. The release is planned for the coming days.

1A Size Record Among Open-Weight Models

Kimi 3 is positioned as the largest open-weight language model from China by parameter count. This matters: open weights mean any company or developer can download and run the model independently, without relying on an API. The company's previous model — Kimi K2 — already posted results competitive with the top benchmarks among open systems.

2Context: Moonshot AI's Rise

Alongside the release preparations, Moonshot AI is raising a new funding round at a $31.5 billion valuation — a significant jump from $20 billion in May 2026. This reflects broad investor appetite for open Chinese models, as their global market share has grown: according to OpenRouter, Chinese models' share rose from 11% a year ago to a peak of 46%.

3What This Changes for the Industry

Models with trillions of parameters existed only inside large closed labs just a year ago. When a system like this is released publicly, the cost of accessing that level of compute drops dramatically. Companies currently paying for Anthropic Opus 4.8 or GPT-5 gain a free alternative with comparable capabilities — and that reshapes the economics of AI development across the entire market.

The release of Kimi 3 continues a trend that began with DeepSeek R1 in early 2025: Chinese open models are systematically closing the gap with American closed systems, shifting the balance of power across the AI industry.