Moonshot AI's Record $35B Valuation and Kimi K3: How a 2.8-Trillion-Parameter Open-Weight Model Is Reshaping the Global AI Race

This was the company's third fundraising tranche in just six months. Moonshot's valuation rocketed from an estimated $4.3 billion in early 2026 to $35 billion by late July, reflecting investor frenzy following the Kimi K3 launch . Annual recurring revenue reportedly reached $300 million in June 2026, up from $200 million in April, while daily sales rose at least sixfold after K3's launch .

Kimi K3: The World's Largest Open-Weight AI Model

Moonshot unveiled Kimi K3 on July 16, 2026, describing it as a sparse Mixture-of-Experts (MoE) model with 2.8 trillion total parameters — making it the largest open-weight AI system ever released . The company committed to releasing full open-source weights on July 27, 2026, which it followed through on as promised .

Key technical specifications:

• Architecture: Sparse MoE with 896 experts, activating only 16 per token (approximately 104 billion active parameters per inference)
• Context window: 1,048,576 tokens (1 million)
• Capabilities: Native vision (multimodal), always-on reasoning mode, OpenAI-compatible API
• License: Modified MIT license, open for research and personal use with business guardrails

Kimi K3 is built on two new architectural innovations: Kimi Delta Attention, which enables faster long-context processing, and a new post-training pipeline that enhanced the model's agentic capabilities .

Competitive Positioning vs. U.S. Models

Moonshot itself stated that Kimi K3 still trails Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol on overall performance benchmarks. However, the company reports that K3 consistently beats Claude Opus 4.8 and GPT 5.5 across coding, reasoning, and knowledge work tests . Some independent analyses described it as performing "neck-and-neck" with top proprietary U.S. systems at a fraction of the cost .

On the BrowseComp benchmark, Kimi K3 scored 91.2, leading in agentic tasks and information retrieval, and ranking second on the AA-briefcase benchmark behind only Claude Fable 5 . It also scored 57 on the Intelligence Index, close to the top closed models .

Pricing: $15 per Million Output Tokens

Kimi K3 API pricing is set at:

• $3.00 per million uncached input tokens
• $0.30 per million cached input tokens (a 90% discount on cache hits)
• $15.00 per million output tokens

These rates are flat across the full 1 million-token context window with no length tiering . By comparison, the estimated cost per task was $0.94 for K3, just under GPT 5.6 Sol's $1.04 and half of Claude Opus 4.8's $1.80 . Kimi K3's API pricing is roughly 30% of Claude Fable 5's rate for a comparable output .

Note that reasoning tokens are billed as output tokens at $15/M, meaning a model's internal chain-of-thought can significantly increase the final bill even for short visible answers .

U.S. Companies Are Already Using Moonshot's Models

Despite geopolitical tensions, U.S. companies have integrated Moonshot's models into production:

• Cursor: The AI coding startup acknowledged it built its Composer 2 feature starting from a Kimi K2.5 base. Cursor's VP of engineering stated that only about 1/4 of the compute spent on the final model came from the base, with the rest from their own training .
• DoorDash: CTO Andy Fang confirmed the company routes lower-level reasoning tasks through Kimi models while reserving Anthropic's Fable for higher-level functions .
• Coinbase: Confirmed internal use of Kimi models .
• Thinking Machines Lab: The AI startup founded by former OpenAI CTO Mira Murati used Kimi while creating its first tool, Inkling .

Data from OpenRouter, an API marketplace, shows that Chinese AI models now account for approximately 60% of token usage by U.S. companies on the platform .

Capacity Crisis: New Subscriptions Paused

Overwhelming demand after K3's launch pushed Moonshot's infrastructure close to its limits. On July 20, 2026, just four days after launch, the company paused new subscriptions for the Kimi chatbot to stabilize service for existing users .

"Kimi K3 has received far more love than we expected," Moonshot wrote in a post on X. "Over the past 48 hours, demand has pushed close to the limits of our current capacity." The company cited "unprecedented compute challenges" and said it would focus on meeting the needs of its current paying customers .

Next Round at $50 Billion Pre-Money and Hong Kong IPO

Even before closing the $3.5 billion round, Moonshot began approaching investors about a follow-on round at a $50 billion pre-money valuation, targeting a final pre-IPO raise in August 2026 ahead of a planned Hong Kong listing . The company aims to complete its IPO on the Hong Kong Stock Exchange as soon as possible within 2026 .

Geopolitical Tensions and Distillation Allegations

Kimi K3's release intensified scrutiny over U.S. AI export controls. The model's performance — achieved despite U.S. semiconductor restrictions — triggered debate about whether China is closing the gap through legitimate innovation or via model distillation from American frontier models .

Multiple reports noted that K3's debut "sent ripples through Silicon Valley" and contributed to a sell-off in Nvidia shares over concerns that cheaper open-weight Chinese models could reduce demand for premium U.S. hardware . The Biden administration faced renewed pressure to tighten export rules on advanced chips and model weights .

Huang Zixin, Moonshot AI's head of enterprise business, denied the distillation accusations in an interview with state media .

The Bigger Picture: What Kimi K3 Means for AI

Kimi K3's release signals a convergence toward open-weight models in the AI industry . By releasing the largest open-weight model ever as a freely downloadable package, Moonshot has given developers worldwide access to frontier-level AI capabilities at a fraction of the cost of comparable proprietary systems .

For builders and enterprises, the message is clear: Chinese AI is now cheap enough and good enough that even U.S. startups and Fortune 500 companies are quietly plugging these models into production to rein in spiraling AI budgets . The question is no longer whether Chinese open-weight models can compete — it is whether proprietary American models can maintain their premium pricing in the face of an open alternative that is "neck-and-neck" on quality.