Zhipu and MiniMax Stocks Crater After Kimi K3 — What Investors Just Priced In
Moonshot’s Kimi K3 — a 2.8-trillion-parameter open-weight model — sent Z.ai stock down 28% and MiniMax down 16% in a single Hong Kong trading session.
On July 17, 2026, Moonshot AI announced Kimi K3 — a 2.8-trillion-parameter model it describes as the world’s largest open-weight AI system — and Hong Kong’s AI stocks immediately repriced. Shares of domestic AI competitors Zhipu (2513.HK) and MiniMax (0100.HK) dropped 27.7% and 16.5%, respectively, in Hong Kong just before market close. The message from institutional investors was blunt: one announcement stripped roughly a quarter of Zhipu’s market value in an afternoon.
Traders drew immediate comparisons to the so-called DeepSeek moment of 2025, when a Chinese lab released a model that performed comparably to US rivals at a fraction of the cost. This time, the disruption is homegrown — one Chinese AI lab undercutting the valuations of two others. That’s a different kind of market signal.
What Kimi K3 Actually Is
Kimi K3 features 2.8 trillion parameters and a 1-million-token context window, using a Mixture-of-Experts architecture with open weights. The sheer scale is the headline: Kimi K3 is a significant jump from its predecessor — K2 had 1 trillion parameters. To put that in domestic context, Kimi K3 dwarfs Chinese competitors including DeepSeek’s V4 Pro (1.6 trillion parameters) and Zhipu AI’s GLM 5 series (744 billion parameters).
Bigger doesn’t automatically mean better — but the benchmarks are hard to dismiss. Arena blind evaluations showed developers choosing Kimi K3 ahead of all top US models on front-end coding tasks, and the model beats Claude Opus 4.8 and GPT 5.5 across coding and general agent evaluations — while placing below Anthropic’s Claude Fable 5 and OpenAI’s GPT 5.6 Sol in overall rankings. According to VentureBeat, on GDPval-AA v2, a benchmark measuring real-world tasks across 44 occupations and 9 major industries, Kimi K3 scored 1,687 — placing it third overall, behind only Claude Fable 5 Max and GPT-5.6 Sol Max.
The architecture matters because it’s the economic argument. Kimi K3 is a sparse mixture-of-experts model, meaning only a fraction of its 2.8 trillion parameters activate on any given request, reducing inference costs compared to a dense model of equivalent size. That’s how Moonshot can price the API at $3 per million input tokens and $15 per million output tokens — the highest pricing from any Chinese AI lab, but roughly half the per-task cost of Anthropic’s Opus 4.8.
The full model weights aren’t out yet. Moonshot has set July 27 as the date for publishing K3’s full weights, leaving developers without the ability to self-host or alter the model in the meantime. The model is available now through Kimi.com, mobile apps, and its API.
Why This Hit Zhipu and MiniMax So Hard
The selloff wasn’t random panic. BofA Securities highlights the fleeting nature of AI leadership, noting increased competitive pressure on firms like Zhipu and MiniMax. According to CNBC, Bank of America analyst Alex Liu wrote that “K3 raises the capability ceiling for China AI models, shifting the burden of proof to other independent AI labs.”
The MiniMax situation is particularly awkward. Just nine days before Kimi K3 landed, reports emerged that MiniMax is building a 2.7-trillion-parameter model — six times the size of its current flagship and the largest in China — which it plans to open-source as early as Q3. Internally it goes by M3 Pro. Moonshot just beat them to the punch, by 100 billion parameters, and by potentially an entire quarter. MiniMax’s big reveal is now a catch-up story before it even ships.
Kimi K3’s 1M-token context and “near-frontier” claims compress differentiation, likely keeping pressure on margins and customer switching costs. When a competitor open-sources a model larger than your flagship-in-progress, your stock taking a 16% haircut is arguably the optimistic reaction.
For Zhipu, the pain is structural. Companies including Moonshot, Zhipu, and MiniMax are releasing powerful models at sharply lower costs, directly challenging Western assumptions that Chinese developers trail their American peers by months. Zhipu was that challenger just weeks ago — now it’s the one being challenged. Zhipu shares lost as much as 30% of their value in Hong Kong trading — the largest single-day decline the company has recorded since its January listing.
The Cost Argument — With an Asterisk
Omdia chief analyst Lian Jye Su noted that Chinese models “can be run at a fraction of the cost that US AI charges its clients,” but added that massive scale “doesn’t necessarily mean you have the best performance by default.” That’s the critical nuance the stock market tends to paper over in its first reaction.
There’s also a hardware wall. As Ryan Fedasiuk of the American Enterprise Institute pointed out, running a 2.8-trillion-parameter model locally requires hundreds of thousands of dollars of computing equipment. Open-weight doesn’t mean free. The MoE architecture keeps API costs reasonable, but self-hosting K3 is not a project for startups running on a few H100s.
And the benchmarks themselves carry a caveat: every published K3 number is a claim made by Moonshot or drawn from API access and can’t be verified until the weights are made public on July 27. Independent stress-testing starts in ten days.
Geopolitics on the Sidelines — and Not So Sidelines
The release coincided with Chinese President Xi Jinping’s remarks at the 2026 World Artificial Intelligence Conference, where he said AI development should not be a “solo performance” by one country. The timing was not subtle.
As the performance gap with US rivals narrows, combined with cost advantages, Chinese AI models are gradually gaining international market attention — and US lawmakers are considering restricting domestic companies from adopting Chinese AI models. Meanwhile, the K3 launch arrives a month after the US government abruptly withdrew Anthropic’s Fable and Mythos models due to security concerns, leaving a gap in the frontier model market that open-weight Chinese models are more than happy to fill.
The broader geopolitical picture is where things get genuinely interesting for investors. K3’s performance underscores a recurring pattern: three years of escalating restrictions on GPUs and lithography equipment have not prevented Chinese labs from reaching or nearing the frontier.
What’s Next for the Losers — and the Winner
Moonshot itself is on a fundraising sprint. The Kimi chatbot developer is seeking as much as $2 billion in fresh funding at a possible $30 billion valuation, launching its third financing effort in six months. For context, Moonshot was valued at $4.3 billion in December 2024 — if the new round closes at $30 billion, it will have multiplied its valuation more than sevenfold in 18 months. According to TechCrunch, Moonshot’s annual recurring revenue topped $200 million in April, driven by rapid growth in paid subscriptions and API usage.
For Zhipu and MiniMax, the path forward is narrower than it was 24 hours ago. MiniMax needs M3 Pro to ship fast and perform better than Kimi K3 — not just match it. The key risk for investors: Zhipu and MiniMax quickly match Kimi K3’s performance and productize faster, reversing the narrative and stopping the de-rating. That’s possible. But the window is closing, and Moonshot just made everyone else’s roadmap look like a work-in-progress.
China’s frontier AI race has officially entered its consolidation phase. The question is no longer whether Chinese labs can compete with the US — it’s which Chinese lab survives to do it.
