The latest
Moonshot has released Kimi K3, China’s largest artificial intelligence model to date, in a sign that the technological gap between Chinese developers and leading US laboratories is narrowing.
The model has 2.8 trillion parameters and will be available for download with open weights, allowing developers and companies to run and modify it. Moonshot’s published results show K3 outperforming Claude Opus 4.8 and GPT-5.5 on most coding and agent benchmarks, although it continues to trail Anthropic’s Fable 5 on most evaluations. The results have not yet been independently verified.
The launch could challenge the view that Chinese models remain eight to 12 months behind their US counterparts. K3’s open-weight structure and lower costs may also offer businesses an alternative to increasingly expensive closed models from US developers.
Details
- Scale: Kimi K3 contains 2.8 trillion parameters, compared with estimates of 1.5 trillion to 2 trillion for Claude Opus 4.8.
- Access: The model will be released with open weights, allowing users to download and customize it.
- Cost: Moonshot’s earlier Kimi K2.6 model costs roughly one-third as much as Anthropic’s Opus 4.8.
- Funding: Moonshot is raising capital at a proposed valuation of about $31.5 billion.
- Valuation gap: Anthropic was valued at $965 billion in May, while OpenAI’s latest valuation reached $852 billion.
Between the lines
K3 does not need to become the world’s highest-scoring model to disrupt the market. Near-frontier performance combined with open access and lower costs could be enough to attract developers seeking to reduce spending and dependence on US laboratories.
What’s next
Independent testing will determine whether Kimi K3 can reproduce Moonshot’s reported results. Its real-world performance, operating costs and adoption beyond China will be the key measures of its impact.
