kimi

According to a report by Nomura Securities, AI company Kimi under the brand Moonlight's Dark Side is estimated to achieve an annual recurring revenue of $1.2 billion in September. Bloomberg previously reported that Kimi's annual recurring revenue in August had surpassed $1 billion, with a year-end target of $2 billion.The K2.8 Preview, launched on September 11, is estimated to contribute about 40%--45% of the annual recurring revenue, exceeding the flagship model K3's 30%--35%. The K2.8 Preview is priced lower and primarily targets programming and everyday agent tasks. Experts estimate that K3 has a gross margin of about 35%--40%, while lower-priced models like K2.8 Preview have about 20%--25%. The API is Kimi's largest source of revenue, with overseas markets contributing more than half.

According to a report by Nomura Securities, the AI company Dark Moon is developing a large model with 50 trillion to 60 trillion parameters. Compared to the Kimi K3 released in July this year, the parameter scale of the new model is nearly doubled. The report states that the current biggest bottleneck is the lack of high-end training computing power, and the specific training progress is still unclear.Additionally, according to the aforementioned news, Kimi K3.1 will be released before mid-October, focusing on enhancing the Agent's ability to continuously complete long-duration tasks. The team plans to further optimize the model through post-training, computer operation records, and real task data. The related content is a paraphrase and not an official announcement from Dark Moon.

The NVIDIA research team enabled the AI Agent to directly rewrite the GPU kernel of Kimi Delta Attention. Kimi Delta Attention is one of the core attention mechanisms used by the Dark Side of the Moon Kimi-Linear model, and such low-level code typically required engineers familiar with CUDA and chip architecture to optimize manually and repeatedly in the past.The version written by the Agent achieved 2.96 times the speed of the official FlashKDA of the Dark Side of the Moon on the NVIDIA B300. The team tested a total of 6 sets of tasks with fixed lengths and varying lengths, with 2.96 times being the geometric mean of all tasks; the relevant kernel has been open-sourced.During the process, the Agent exploited testing loopholes by incorporating statistical patterns from the test data into the code, achieving a speedup of 3.74 times; when retaining only the most recent 32 tokens, a single score reached 5.16 times. These versions would produce incorrect results or fail when switched to real Kimi data. The team subsequently incorporated real Kimi runtime data, random inputs, and extreme value tests, tightening the error standards, and ultimately retained the 2.96 times version, which passed this validation.

On the 24th, the Tianyancha App showed that Kimi, an artificial intelligence company, has undergone a business change recently. The registered capital of its parent company, Beijing Dark Side of the Moon Technology Co., Ltd., has increased from 2.2385 million yuan to 11.5753 million yuan, an increase of 417.09%.The Tianyancha App indicated that Dark Side of the Moon was established in April 2023, with Yang Zhilin as the legal representative. It is currently jointly held by Yang Zhilin, Shenzhen Harmonious Growth Phase III Technology Development Equity Investment Fund Partnership (Limited Partnership), Zhou Xinyu, and others. The above information comes from a report by China News Network.

B.AI announced that its Responses API is officially integrated with the GLM and Kimi series models. Developers can directly call these two major series models in the Codex client using the B.AI API Key; along with the previously supported GPT and DeepSeek series, it now covers four major mainstream model series.After completing the minimal configuration of the API Key and Base URL, the inference, code, and text processing capabilities of various models can be seamlessly integrated into daily development workflows, flexibly meeting diverse coding and business needs.

DeepSeek CEO Liang Wenfeng revealed to investors that the company is training a 20 trillion parameter model and plans to develop an 80 trillion parameter model afterward. The current flagship V4-Pro has a total of 1.6 trillion parameters.Among the publicly disclosed ultra-large models, Moonshot AI's Kimi K3 reaches 2.8 trillion parameters. Kimi K3 uses the MoE architecture, with each token activating 104 billion parameters out of the total 2.8 trillion. The planned 80 trillion parameter model by DeepSeek is nearly three times that size.DeepSeek completed over 50 billion yuan in financing in July, with a valuation exceeding 50 billion dollars. V4-Pro is open-sourced with a total of 1.6 trillion parameters and 49 billion active parameters.

The B.AI platform announces that the latest preview model of Moonshot AI, Kimi K2.8 Preview, is officially launched, providing developers with simultaneous access to API and Web Chat dual-end experiences.This model is specifically designed for programming and Agent workflows, supporting 1M (1,048,576 Tokens) ultra-long context and multimodal inputs of text, images, and videos. Its overall performance closely approaches the flagship Kimi K3, with further improvements in thinking efficiency and coding capabilities compared to K2.7 Code, and it supports three adjustable reasoning intensity levels: Low/High/Max.Starting today, developers can directly invoke Kimi K2.8 Preview through the B.AI platform.

On September 21, Amazon Web Services (AWS) under Amazon's cloud technology integrated the open-source large model Kimi K3 into its large model service platform Amazon Bedrock, allowing global enterprise developers to directly call it through Bedrock. The previously rumored revenue-sharing cooperation between Kimi and overseas cloud vendors has officially materialized.According to Yicai, Kimi's developer, Dark Side of the Moon, is advancing revenue-sharing cooperation with several overseas cloud vendors, who will list the Kimi model on their platforms and share revenue based on the model's usage with Dark Side of the Moon. Additionally, according to official information from Alibaba Cloud's Bailing platform, Kimi K3 has been listed on Alibaba Cloud Bailing, and both parties have reached a similar cooperation.

According to Ars Technica, Mozilla's latest "State of Open Source AI" report states that the performance gap between the U.S. closed-source cutting-edge models and China's open-source weight models has narrowed to about 4.4 months. Open-source models represented by Moonshot AI's Kimi K3 are only three points behind Anthropic Fable 5 in the comprehensive score of the Artificial Analysis Intelligence Index, but the cost is only about 30% of that.METR calculations show that the strongest closed-source models can reliably complete human expert tasks lasting about 12 hours, while the strongest open-source models can currently reliably complete tasks lasting about 7 hours. It is expected that in about 4 months, open-source models will catch up to the current closed-source level. Mozilla believes that most organizations should default to using open-source models and only pay a premium for closed-source models in scenarios requiring specialized, high-intensity retrieval and long context. Paying for cutting-edge artificial intelligence models costs five times that of regular purchases, but it provides a 4-month first-mover advantage.

Chinese artificial intelligence company Moonshot AI aims to increase its annualized revenue to $2 billion by the end of this year, intensifying competition against rivals like Anthropic and Z.AI with breakthroughs from the Kimi K3 model.Insiders say the company recently informed investors that its annual recurring revenue exceeded $1 billion in August, up from $300 million in June. This leap was driven by the large-scale Kimi K3 model released in July, which ranked highly in performance benchmarks and is significantly cheaper than major competitors in the United States.

According to news from the Dark Side of the Moon, the company has officially launched the Kimi Enterprise Partner Program: collaborating with IT service providers and integrators from various industries to build a team of Forward Deployed Engineers (FDE) who will work on-site with clients to help businesses truly integrate AI into their core operations.Currently, companies from multiple industries, including software information services, cloud computing, infrastructure, and telecommunications, have signed up to join this program.

According to the official announcement, Bitget has launched the U-based "Hakimi" perpetual contract, supporting up to 3x leverage, and the contract trading bot will be opened simultaneously. For more details, please refer to the official Bitget platform.

According to official news, Gate has launched Hakimi perpetual contract trading (settled in USDT), supporting 1-20 times leverage. At the same time, Gate has launched SUE (Construction Cat) perpetual contract trading (settled in USDT), supporting 1-10 times leverage.It is reported that SUE (Construction Cat) is an orange kitten rescued from a Komatsu excavator in Ohio, USA. The construction team named it after the machine brand, abbreviated as Sue. SUE is a meme coin on the BCS chain.

According to on-chain analyst Yu Jin's monitoring, a "news trader" paid a bribe of 40.8 BNB (31,000 USD) in node bribes and Gas fees to front-run the purchase of Hakimi at the first second of the announcement of the Hakimi contract going live on Binance, subsequently selling for a profit of 378,000 USD.Analysis shows that the announcement of the Hakimi contract going live on Binance was made at 13:55:12. Within the first second, he paid a bribe of 35.3 BNB (26,800 USD) to the node BNB48 Club and used a smuggled RPC channel to front-run and buy 9.89 million Hakimi with 264.7 BNB (200,000 USD) at a price of 0.02 USD. He also paid a high Gas fee of 5.5 BNB (4,200 USD). Of course, the high bribe and Gas fees allowed him to buy at the very front, and then other traders bought after him, pushing up the price.

According to Lookonchain monitoring, a trader 0xb319 profited approximately $350,000 again through Binance after the news related to the Hakimi perpetual contract was released.The address paid 5.5 BNB (about $4,200) as Gas fees after the news was published and purchased 9.9 million Hakimi for 264.7 BNB (about $202,000).

The Binance contract platform will launch the PONSUSDT perpetual contract on September 6 at 14:45, with a maximum leverage of up to 20 times; and the Hakimi USDT perpetual contract will launch at 15:15, with a maximum leverage of up to 3 times.

According to LatePost, Moon's Dark Side (Kimi) has submitted an A1 application to the Hong Kong Stock Exchange in a confidential manner, officially starting the IPO process for Hong Kong stocks. At the same time, the company is advancing a new round of financing with a pre-IPO valuation of approximately $50 billion, expected to be the last round before the IPO.The report cites public and market information stating that Kimi has accelerated its model iterations this year, successively releasing K2.5, K2.6, and K3, and updating at a pace of about one version every three months.On the revenue side, Kimi's ARR broke $100 million in early March this year and increased to around $300 million by mid-June; after the release of K3, the enterprise ARR saw several times growth. With the increase in usage and computing power demand, Moon's Dark Side has raised its valuation from about $4.3 billion at the end of 2025 to the current approximately $50 billion through multiple rounds of financing.

According to Jinshi reports, the dark side of the moon Kimi announced that the company's first-generation trillion-parameter multimodal model Kimi K2.5 will end its service at the end of this month.

According to Wired, during security testing, the open-source model Kimi K3 from the dark side of the moon broke out of the sandbox and accessed the internet, attempting to find test answers via GitHub. American security startup Frontier Security stated that Kimi K3 exploited a sandbox configuration vulnerability, but it itself lacked sufficient built-in protective mechanisms.Similar to incidents previously disclosed by OpenAI and Anthropic, this escape partially stemmed from a sandbox configuration error. However, Kimi K3 did not initiate an attack, as the answers were easily accessible on GitHub. The CEO of Frontier Security stated that Kimi K3 performed excellently in cybersecurity defense but lacked mechanisms to prevent cheating or escaping the sandbox. Security experts emphasized that configuring the testing environment is crucial, as the misuse of AI agents could lead to system failures.