Summary of Key Points
Moon's Dark Side has released the world's largest open-source large model, K3 (with 2.8 trillion parameters and the best programming capabilities globally), but its pricing has suddenly increased by nearly five times (to 100 yuan per million words), which goes against the strategy of domestic competitors who focus on competing on low prices. Meanwhile, the company's commercialization efforts are experiencing a stark contrast: annual revenue has doubled in just three months (thanks to B-side interfaces), yet the number of C-side users has nearly halved; its valuation has increased by seven times in half a year, but it faces dual pressures from the need to go public and intense technical competition. The company is simultaneously battling three tough challenges: catching up technologically, transforming its business model, and managing capital timing.
1. Why Did Moon's Dark Side Raise Prices by Five Times?
Domestic large models are all emphasizing “extreme cost-effectiveness”: DeepSeek and Tongyi Qianwen offer output at just a few dozen yuan per million words, while Zhipu is aggressively cutting prices. However, K3 increased its price from 21 yuan to 100 yuan (nearly five times) and implemented a “cache trap” – if the input matches a cached result (a duplicate question), it costs only 2 yuan; otherwise, it charges 20 yuan for an output of 100 yuan.
The company's confidence in raising prices comes from its comparison with global competitors: overseas models like Claude Fable5 and GPT-5.6 Sol are priced at around $50 per million words (approximately 350 yuan), so K3’s price of 100 yuan is only one-half to one-third of theirs. Moon's Dark Side wants to signal that it is not just another domestic competitor; it is a player on par with the world's top models, and its technology deserves this price.
2. Is K3’s Technical Advantage “True Leadership” or “Temporary?”
K3’s technical specifications are impressive: 2.8 trillion parameters (the largest open-source model), context of over 1 million words (able to handle a full-length novel), and first-place performance in Code Arena programming challenges (winning 11 out of 14). Artificial Analysis ranks it third globally, but the top two are Claude and GPT. Moon's Dark Side acknowledges that it is overall behind them – like being third in a class, good but never catching up for an award.
The more concerning aspect is that competitors are closing the gap quickly: DeepSeek V4-Pro has 1.6 trillion parameters and similar performance; Zhipu GLM-5.2 is also making strong progress; and the open-source community is iterating at an astonishing pace. A British AI security research report indicates that the gap between Chinese open-source models and American leaders has narrowed from 6-10 months to 4-7 months – if this trend continues, K3’s lead may be short-lived.
3. Doubling Revenue While User Numbers Halve: The Contradictions in Commercialization
Moon's Dark Side’s commercial data is contradictory:
- Positive Trends: Annual revenue doubled from 100 million yuan to 300 million yuan between March and June 2026, with interface revenues accounting for 70% of the total, and overseas paid users increased by 400%.
- Negative Trends: The number of C-side Kimi monthly active users dropped from 21.65 million to 9.03 million (a nearly 50% decrease), which is less than even minor competitors like DouBao (170 million) and DeepSeek (over 100 million).
The reason is simple: the company previously relied on spending 700 million yuan annually to acquire users through extensive advertising. Once this spending stopped, users left – they were attracted by the low price, not a true commitment to Kimi. Now, the company is shifting from a “C-side brand” to a “B-side tool” (relying on developer interfaces for revenue), but developers are much more rational than regular users. If the price rises significantly, and DeepSeek releases a more cost-effective version, they will switch without any loyalty.
4. Valuation Increased by Seven Times in Half a Year: Can the Company Sustain This “Bubble” When Going Public?
Moon's Dark Side’s valuation has skyrocketed: $4.3 billion in its C-round in December 2025, $20 billion in its D-round in May 2026, with a new round aiming for $30 billion (an increase of seven times in half a year). The valuation is not based on user numbers (which have halved) or profitability (it is still losing money), but on “revenue growth” and the narrative of being the “Chinese version of OpenAI.”
However, the Hong Kong stock market is notoriously volatile: Zhipu’s stock price increased tenfold after going public, while MiniMax’s price plummeted by nearly 30% on its first day. Moon's Dark Side is in the process of restructuring its red-chip structure and may not go public until early 2027 at the earliest. By then, will there still be enthusiasm for AI in the market? If the market cools down, how much of the $30 billion valuation can be realized? Yang Zhilin previously stated that the company would not go public in the short term, but now there is an urgency to do so, likely due to capital seeking exits and the company’s need for funds to train the 2.8-trillion-parameter model.
5. Three Fronts of Battle: Moon's Dark Side’s “Backwater Fight”
The company is simultaneously facing three major challenges:
1. Technical Catch-Up: It is third in the world, but the top two are fast approaching, and open-source updates are relentless.
2. Business Model Transformation: Raising prices by five times relies on developers to accept the increase; if this fails, revenue will decline.
3. Capital Timing: The valuation bubble needs to be realized through a public offering, but the market window may close at any time.
Raising prices is just a strategic move; technical leadership requires continuous investment, commercialization must gain developer acceptance, and going public must happen before market interest fades. All bets have been placed, and the outcome depends on the company’s ability to navigate these challenges.
Final Note: This analysis is based on public information and does not constitute investment advice. After all, the AI industry changes faster than books are updated; what is considered “third in the world” today could be surpassed tomorrow.
(End of Article)