Moonshot AI is preparing to release Kimi K3, a 2 trillion to 3 trillion parameter open-weight model that the Chinese lab expects to match or surpass Anthropic's Opus 4.8, according to the Financial Times. The model would be the largest open-weight system ever shipped from China and is due 'in the coming days.' Moonshot is also raising fresh capital at a $31.5 billion valuation, up from $20 billion in May.
The valuation jump is roughly 58% in two months. Moonshot raised $2 billion in that May round, and the new mark puts it firmly in the tier of Chinese AI labs that Western investors and enterprise buyers are watching as serious alternatives to closed US frontier models.
Kimi K2, the prior release, already ranked well on public benchmarks and demonstrated capabilities not far off the closed-source frontier. K3 is Moonshot's attempt to close that remaining gap outright, and the parameter count — 2T to 3T — signals the company is scaling rather than trying to squeeze more out of a smaller architecture.
Key facts
- 01Moonshot's Kimi K3 will ship with 2 trillion to 3 trillion parameters, the largest open-weight model ever released from China.
- 02Moonshot is raising at a $31.5B valuation, up from $20B in May when it raised $2B.
- 03Kimi K3 is expected to match or surpass Anthropic's Opus 4.8, per the Financial Times.
- 04The release is expected 'in the coming days,' following Kimi K2's strong benchmark run.
- 05The launch feeds a growing enterprise pitch to swap closed models like ChatGPT and Claude for open weights from DeepSeek, Z.ai, or Moonshot.
The comparison target matters. Anthropic's Opus line sits at the top of Anthropic's own pricing and capability stack, and matching it with open weights would let enterprise customers self-host a model with broadly similar performance instead of paying per-token API fees to a US lab.
That pitch is landing at a useful moment. A fresh debate is running through enterprise AI buyers about whether to keep paying labs like OpenAI and Anthropic for expensive closed models, with executives raising concerns that data submitted to products like ChatGPT and Claude could be retained or used in ways clients cannot fully audit.
The alternative being pitched by rival vendors is straightforward: take a cheaper open-weight model from DeepSeek, Z.ai, or Moonshot, fine-tune it on internal data, and run it on infrastructure the buyer controls. Kimi K3 slots into that pitch as the highest-parameter option on the menu.
Moonshot's rise sits alongside DeepSeek and Z.ai as the three Chinese labs that Western competitors now name-check most often. DeepSeek's releases have repeatedly forced pricing conversations at US labs, and Moonshot's K2 built a following among developers running local or self-hosted inference stacks.
The caveats are worth stating plainly. Parameter count is not capability — a 3 trillion parameter model can still underperform a smaller, better-trained one, and independent benchmark results for K3 do not yet exist. Financial Times' reporting relies on anonymous sources, and Moonshot has not published official specs or a release date. Opus 4.8 itself is a moving target, with Anthropic iterating quickly.
For the open-weight market, Moonshot pushing to Opus-tier performance would compress the gap that has defined AI pricing power for the past two years. Closed labs charge a premium because their models are meaningfully better; if that premium narrows to the point where enterprise buyers can rationally choose a fine-tuned open model for most workloads, the revenue math for OpenAI and Anthropic on general-purpose API traffic gets harder, and the pressure to differentiate on agents, tools, and vertical products intensifies. A $31.5B valuation on a Chinese open-weight lab is the market pricing that shift in.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
