Moonshot AI’s “The Dark Side of the Moon” is set to launch Kimi K3 soon! With 2.8 trillion parameters and million-long context, its performance is close to Anthropic Opus 4.8

The technological gap between China and the United States in cutting-edge artificial intelligence is rapidly shrinking! Based on disclosures from the overseas media TechCrunch and official API documentation, China’s AI unicorn Moonshot AI is about to roll out its most powerful model, “Kimi K3.” The model has up to 2.8 trillion parameters, natively supports visual understanding and an ultra-long context of one million tokens, and its performance is considered to be already close to Anthropic’s Opus 4.8, a leading U.S. frontier institution. Reports say Moonshot AI is carrying out a new round of fundraising, with a valuation possibly reaching $31.5 billion.
(Background: Beijing is rumored to be considering controls on DeepSeek and the Dark Side of the Moon, because China’s AI models are being snapped up cheaply by Silicon Valley)
(Background addition: The Dark Side of the Moon “isn’t in a hurry to go public,” yet it’s talking to Goldman Sachs about a Hong Kong stock IPO—valuation has already climbed to $18 billion)

Table of contents

Toggle

• A 2.8 trillion-parameter beast, native visual and deep reasoning support
• Performance close to Opus 4.8, valuation skyrockets to $31.5 billion
• “Open-weight” becomes a strategic weapon to counter closed-source giants

The AI technical landscape between China and the United States—at the cutting edge of artificial intelligence (AI)—is now set to undergo a major shift in its segment lineup. On July 16, 2026, Taipei time, citing a report by the Financial Times, TechCrunch reported that China’s AI startup Moonshot AI’s latest-generation large language model, “Kimi K3,” is expected to be formally announced within the next few days.

In fact, on Moonshot’s official Kimi API platform, a quickstart guide titled “Kimi K3 Quickstart” has already quietly gone live, giving developers an early look at the powerful specs of this new flagship model.

A 2.8 trillion-parameter beast, native visual and deep reasoning support

According to the official released documents, Kimi K3 has a parameter scale of up to 2.8 trillion (2.8 trillion parameters). This will make it, all at once, the largest open-weight (open-weight) AI model in China to date. In terms of underlying architecture, K3 adopts an innovative hybrid linear attention mechanism (Kimi Delta Attention). Not only does it natively support an ultra-long context window of up to 1,000,000 tokens, it also adds strong visual understanding capabilities.

It’s worth noting that in its developer guide, the official specifically emphasizes that K3 defaults to enabling an extremely deep “thinking mode.” Currently, its default reasoning effort (reasoning effort) is set to the highest level, and the system will automatically return detailed reasoning content (reasoning content) along with the final result. This is especially suitable for advanced knowledge-work scenarios such as software engineering and complex mathematical proofs (such as proving that sqrt(2) is irrational).

Performance close to Opus 4.8, valuation skyrockets to $31.5 billion

In terms of performance, industry insiders generally speculate that Anthropic’s target model, Opus 4.8, has approximately 1.5 trillion to 2 trillion parameters. And TechCrunch’s report says that with its larger parameter scale, Kimi K3’s overall capabilities are expected to be on par with Opus 4.8, and in some benchmark tests it may even manage to surpass it.

This technical breakthrough has helped Moonshot AI stand out in the capital markets. Earlier this May, the company just completed a $2 billion funding round at a valuation of $20 billion. Now, according to the latest reports, Moonshot AI is currently in a new fundraising round, and it is expected that the company’s overall valuation will surge to $31.5 billion.

“Open-weight” becomes a strategic weapon to counter closed-source giants

The strong rise of Kimi K3 also highlights subtle changes in the current enterprise AI market. The report analyzes that as more and more enterprises worry that uploading confidential business data to closed-source models such as OpenAI or Anthropic could create privacy risks, demand in the market for “open-weight models” with top-tier performance is rising rapidly.

Compared with expensive closed-source solutions, using models such as Kimi K3 allows enterprises to deploy locally and retain control over their data at a lower cost (the official emphasizes low-priced pay-as-you-go, and no tiered billing for long text). This trend not only is forcing tech giants such as Microsoft to start adjusting their sales strategies, but also gives Moonshot AI a chance to secure a more critical strategic position in the global AI ecosystem.