The world's largest open source model is here, Dark Side of the Moon releases Kimi K 3 with 2.8 trillion parameters; Space X cancels the 13 th test flight of "Starship", and the stock price dives by more than 4% after the market
| Friday, July 17, 2026 | The NO.1 world's largest open source model is here. Dark Side of the Moon releases KimiK3 with 2.8 trillion parameters. In the early morning of July 17, Beijing time, Dark Side of the Moon officially released KimiK3, the strongest model to date. This model has 2.8 trillion parameters and 1 million token context windows. It is currently the world's first open source model exceeding 2 trillion levels.
| Friday, July 17, 2026 | NO.1 The world’s largest open source model is here, Dark Side of the Moon releases Kimi K3 with 2.8 trillion parameters In the early morning of July 17th, Beijing time, Dark Side of the Moon officially released Kimi K3, the most powerful model to date. This model has 2.8 trillion parameters and 1 million token context windows. It is currently the world's first open source model exceeding 2 trillion levels. Official evaluation shows that the comprehensive intelligence level of Kimi K3 is second only to Claude Fable 5 and GPT-5.6 Sol, and its performance is particularly outstanding in code benchmark tests such as long-cycle software engineering. In addition, Dark Side of the Moon has completed 6 rounds of financing this year, with its latest pre-investment valuation reaching US$31.5 billion. As of mid-June, its annual recurring revenue has exceeded US$300 million, of which API revenue accounts for more than 70%. Comment: Dark Side of the Moon released the 2.8 trillion parameter open source model Kimi K3, approaching the strength of the world's top closed source models, marking the official entry of domestic AI into a new stage of trillion-level ecological competition. NO.2 SpaceX canceled the 13th test flight mission of "Starship", and the stock price plunged more than 4% after the market closed. On July 16, local time, Musk's space exploration company SpaceX canceled a major test flight of the "Starship" rocket. "Starship" was originally scheduled to take off from SpaceX's "Star Base" launch facility in southern Texas at 5:45 pm local time, performing its 13th major mission, the first launch since the company's U.S. IPO. Musk said that because part of the engine failed to start, triggering an automatic abort of the launch, he is expected to try to launch the rocket again in the next few days. After the news was announced, SpaceX's stock price fell by more than 4% after hours. It has fallen by about one-third since its June high, and its market value has evaporated by more than $860 billion. Comment: The temporary cancellation of the Starship test flight has hit market sentiment, and the previous IPO enthusiasm has cooled down. NO.3 Japanese AI Alliance Noetra plans to purchase 27,500 NVIDIA Rubin GPUs On July 16, local time, Japanese AI company Noetra announced that it would cooperate with NVIDIA to build an AI computing platform, planning to deploy approximately 27,500 NVIDIA Rubin GPUs and put them into operation in June 2028. Noetra was founded by Sony, SoftBank, NEC, Honda and other companies. The goal is to develop multi-modal AI basic models with image, video, audio and physical space understanding capabilities, with a focus on robot AI applications in the future. On the same day, NVIDIA also announced that it would expand the provision of basic robot development model "Cosmos" to Japanese companies to accelerate the construction of Japan's physical AI ecosystem. Comment: Japan is using NVIDIA’s advanced computing power to accelerate the integration and development of robots and AI models, and promote the upgrading of manufacturing industry to intelligence. NO.4 Google Gemini flagship model is delayed, and AI programming capabilities are accused of lagging behind OpenAI On July 15, local time, according to media reports, the release plan of Google’s flagship model Gemini 3.5 Pro was delayed several months from the original time. The main reason was that the company was trying to improve its model programming capabilities, but the progress was not as good as expected. According to reports, the new models recently launched by OpenAI and Meta have further led Google’s existing products in AI code generation capabilities. It is reported that some engineers and researchers within Google are worried that the company is gradually losing its leading edge in the AI competition, and some talents have flowed to competitors such as Anthropic. This Gemini postponement also highlights that the competition for AI models has entered the stage of competition for application capabilities such as code generation and intelligent agents. Comment: The focus of competition is shifting from model parameters to actual application performance, and programming capabilities have become an important direction for large model manufacturers to compete. (