Chinese AI lab Moonshot AI is preparing to release Kimi K3, an open-weight model with 2-3 trillion parameters that aims to match Anthropic's closed-source Opus 4.8 performance.

The model represents a significant leap from Moonshot's current Kimi K2 series, which has already gained traction in open-source AI circles for benchmark performance close to frontier models.

Kimi K3 will be the largest open-weight AI model from China when released "in the coming days," according to Financial Times sources. The model's parameter count places it among the largest publicly available foundation models globally.

Funding and valuation surge

Moonshot is simultaneously raising fresh capital in a round that would value the company at $31.5 billion, up from its $20 billion valuation in May when it raised $2 billion.

The Beijing-based company has positioned itself as a leading alternative to Western AI labs, particularly as enterprises grow concerned about data privacy with closed-source models from OpenAI and Anthropic.

Industry executives increasingly worry that AI labs might extract proprietary data submitted through products like ChatGPT and Claude for their own model training purposes.

Open-source momentum

The release comes as companies explore cheaper alternatives to expensive closed-source models. Open-source options from Chinese labs like DeepSeek, Z.ai, and Moonshot allow enterprises to fine-tune models for specific use cases without ongoing API costs.

Moonshot's Kimi series has already demonstrated strong performance on standard benchmarks, with K2 models ranking competitively against much larger closed-source systems.

The company's long-context capabilities have been particularly notable, with Kimi handling extended conversations and document analysis tasks that rival established Western models.

Kimi K3's release timeline suggests Moonshot aims to capitalize on growing enterprise interest in open-weight alternatives before competitors can respond with similar offerings.