Moonshot claims its new AI model matches performance of OpenAI and Anthropic

A Beijing-based AI startup just dropped what it says is the biggest open-weight AI model on the planet. And the early numbers suggest it’s not bluffing.

Moonshot AI released Kimi K3, a model packing 2.8 trillion parameters and a 1-million-token context window. The company claims performance that rivals or surpasses top-tier offerings from Anthropic and OpenAI, particularly in coding and knowledge-work benchmarks.

What makes Kimi K3 different

Kimi K3 claimed the top spot on Frontend Code Arena with an Elo rating of 1679, and landed second overall on several other independent leaderboards. In coding tasks and GPU optimization specifically, independent evaluations placed it ahead of leading US models.

Moonshot is emphasizing cost advantages, positioning Kimi K3 as a model that delivers frontier-level capabilities without frontier-level pricing.

The company behind the model

Moonshot AI is backed by Alibaba and Tencent. The company is reportedly aiming to raise $2 billion at an estimated valuation of $30 billion ahead of a potential listing on the Hong Kong Stock Exchange.

The company previously released Kimi K2 in July 2025 and followed it with K2 Thinking in November 2025. Kimi K3 launched on July 16, 2026, and its release coincided with notable declines in share prices for other Hong Kong-listed AI firms.

What this means for investors

If Kimi K3’s benchmark claims hold up under broader real-world testing, it strengthens the case that the AI moat for US-based labs is narrower than their valuations suggest. It’s one thing when open-source models trail proprietary ones by a generation. It’s another thing entirely when they’re matching or beating them on key benchmarks while charging less.

Moonshot’s potential $30 billion valuation and Hong Kong IPO trajectory also signals that the center of gravity in AI investment may be shifting eastward. Alibaba and Tencent’s backing gives the company distribution advantages across Asian markets that US competitors would struggle to replicate quickly.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.