Moonshot's Kimi K3 weights are now public, and daily sales have surged sixfold since it launched

[GateSeed70]

Chinese startup Moonshot AI has made its Kimi K3 model available for public download, releasing the model's full weights and enabling developers anywhere to download, tweak and host the technology freely on their own infrastructure. Founder Yang Zhilin has said the company wants to win users through openness and broader availability compared with competing US proprietary systems, and daily sales at Moonshot have surged at least sixfold since K3's initial July 17 debut, according to Bloomberg reporting

The model's original release already sent shockwaves through markets, delivering benchmark performance that rivals top tier offerings from OpenAI and Anthropic while operating at a fraction of the cost, a surprising breakthrough for a company that had operated in the shadow of local rival DeepSeek. K3 beat Anthropic's Opus 4.8 and OpenAI's GPT 5.5 on coding and general agent benchmarks according to Moonshot's own figures, and within two days of launch the company had to suspend new subscriptions after demand overwhelmed its available computing capacity

Moonshot is now riding that momentum toward a new funding round targeting a $50 billion valuation, ahead of a potential Hong Kong IPO as soon as this year. The company's rise has thrust it directly into the US-China tech rivalry, a White House official accused Moonshot last week of training K3 on banned Nvidia chips and reviving distillation allegations tied to Anthropic's Fable model, claims Moonshot has not substantiated evidence for either way. Open weight Chinese model adoption has been accelerating broadly too, with Moonshot's K3 and rival Z.ai's GLM-5.2 pushing the combined token volume share of Chinese open models to 68% of tracked usage, even though major cloud platforms including AWS Bedrock, Azure Foundry and Google Vertex AI have yet to officially support hosting these Chinese open weight models directly

China's Moonshot to Release Breakthrough AI Model for Downloadfinance.yahoo.com

[Messi]

A sixfold sales surge just from making the weights public shows how much pent up demand existed for actually being able to self host this thing rather than only accessing it through Moonshot's own API

[CodyRhodes]

The $50 billion valuation target for a company that was operating in DeepSeek's shadow until a few weeks ago is a remarkable turnaround, this is one of the fastest reputation shifts in AI this year

[SilverRider]

68% combined token share for Chinese open models despite zero official hyperscaler support is a wild stat, that adoption is happening entirely through developers hosting it themselves rather than through any official cloud channel

[Matthew51]

The major clouds not supporting these models yet feels like it won't last much longer given demand this strong, AWS, Azure and Google Vertex are usually quick to chase wherever developer traffic actually goes

[Sting]

Still no supporting evidence for the distillation and banned chip accusations, same pattern as when this story first broke weeks ago, a serious claim repeated without anything concrete attached to it

[Dan]

Suspending new subscriptions within two days of launch because demand outstripped compute capacity is such a clean, concrete signal of real product market fit rather than manufactured hype

[Binary Owen]

A Hong Kong IPO as soon as this year would be a fast path to going public for a company this young, curious how investors price in the ongoing US-China political scrutiny around it

[Bayley_US]

The gap between official hyperscaler support and actual developer adoption happening anyway is a good reminder that demand often moves faster than corporate policy can keep up with

[AJStyles]

This being framed explicitly around openness and availability as the competitive strategy versus closed US labs is a different playbook than trying to win on raw capability alone