The origin story of Moonshot AI is truly unique. Yang Zhilin, Zhou Xinyu, and Wu Yuxin, three friends from Tsinghua University, founded the company in March 2023. Its name, which roughly translates to “Dark Side of the Moon,” was chosen to commemorate the 50th anniversary of Yang Zhilin’s favorite album. For a company that is currently creating models with 2.8 trillion parameters, that detail seems almost too human. Nevertheless, it fits in some way.
Yang’s declared objective has never been very modest. His goal is to develop foundation models that can produce artificial general intelligence. His three milestones—long context windows, a multimodal world model, and a scalable architecture that can self-improve without human input—read like the aspirations of someone who genuinely thinks AGI is feasible rather than merely a helpful story for fundraising. It’s still unclear if that belief is justified. However, the company’s output indicates that they are doing more than just talking.
In October 2023, Moonshot introduced Kimi, its first chatbot. When it first launched, it was capable of processing up to 200,000 Chinese characters in a single conversation, which was a noteworthy technological accomplishment. The company stated that figure had increased to two million by March 2024. The service crashed for two days as a result of the unexpected spike in users that followed. Moonshot apologized in public. It was the kind of escalating discomfort that usually accompanies a product that succeeds.
Money came in fast. In February 2024, Alibaba led a $1 billion fundraising round, obtaining a 36 percent share and valuing the business at $2.5 billion. Later that year, Tencent joined. At a pre-money valuation of $3.8 billion, the company was reportedly nearing an additional $600 million by October 2025, led by IDG Capital. Those figures are startling for a business with about 300 workers. Investors appear to think that something genuine is taking place.
A sort of turning point was the July 2025 release of Kimi K2. It was a serious open-source model with one trillion total parameters and 32 billion active during inference. The day after it was released, it was the most downloaded model on Hugging Face. Going open-source seemed like a conscious choice. Chinese AI firms have been moving toward transparency, in part as a calculated response to American export restrictions that restrict access to cutting-edge chips and technology. It will be interesting to see if this transparency is tactical or long-lasting.
The Kimi K3, which debuted in July 2026 and has 2.8 trillion parameters, is the biggest open-weight model that has ever been made public. On benchmarks, it performed better than Claude Opus 4.8 and GPT-5.5, only falling short of Claude Fable 5 and GPT-5.6 Sol, all while being reasonably priced. It obtained an Elo of 1547 on long-horizon knowledge work evaluations. According to reports, the model edited its own promotional video using 56 source clips, managing beat synchronization and several rounds of editing on its own. Compared to a benchmark number, that detail is more difficult to ignore.
There has been some controversy surrounding the company. Moonshot was accused by Anthropic in February 2026 of creating millions of conversations with Claude using thousands of fictitious accounts in order to train its own models. The alleged endeavor was characterized by a White House adviser as widespread theft of American AI capabilities. The accusations have not been fully acknowledged by Moonshot in public. Although the legal and diplomatic ramifications of this are still unknown, the accusations gave an already complex geopolitical story a sharp edge.
Observing Moonshot’s development over the last three years gives the impression that something truly competitive has emerged—not a replica of Western AI, but a business with its own infrastructure, research directions, and subtly expressed belief that AGI is worthwhile. The dark side of the moon was always present, regardless of whether the name has any significance beyond branding. To see it, you simply needed the proper gear.
