Blab AI
Moonshot AI Unveils 2.8-Trillion-Parameter Open Model Kimi 3, Prompting Market Shock
On July 16, 2026, Beijing‑based Moonshot AI—one of China’s six AI Tigers—announced the launch of Kimi 3, a large language model that pushes the envelope with 2.8 trillion parameters, a 1 million‑token context window, native multimodality, and a mixture‑of‑experts (MoE) architecture. The unveiling coincided with a brief dip in the Nasdaq Composite, sparking commentary from analysts and investors alike.
Founded by drummer Yang Zhilin, a Pink Floyd aficionado, the company’s name—Yuezhi Anmian in Chinese, literally “Dark Side of the Moon”—reflects his musical passion. The international brand, Moonshot, signals an ambition to reach new frontiers. The firm has positioned itself as a challenger to U.S. incumbents such as OpenAI and Anthropic.
Kimi 3 is the first open‑weight model to cross the 2.8‑trillion‑parameter threshold. A post on X announced that the platform’s API and products now grant access to the model, while the full set of weights is slated for release on July 27, 2026. The MoE design activates only a subset of parameters for each inference, improving efficiency for long‑horizon agentic coding tasks. Its multimodal capabilities let it process visual inputs natively, and the 1 million‑token window expands the length of text it can handle in a single prompt.
The announcement places Kimi 3 in direct competition with Anthropic’s Claude 5 and OpenAI’s most advanced closed models. Moonshot AI claims the new model rivals U.S. leaders on several performance benchmarks and notes a pricing structure that undercuts most competitors. The open‑weight nature of Kimi 3 is significant because it allows developers to download, modify, and deploy the model on their own infrastructure—an option unavailable for the current generation of U.S. models.
Market reaction was swift. Within hours of the announcement, the Nasdaq Composite fell, reflecting investor concern over a potential shift in the competitive landscape of large language models. Analysts noted that the release of an open‑weight, 3‑trillion‑parameter model from China could accelerate the spread of advanced AI capabilities beyond the U.S. cloud ecosystem.
Moonshot AI’s strategy to release the full weights a month after the API launch is intended to balance early commercial use with broader community engagement. The company has already positioned Kimi 3 as a professional coding tool, citing lower pricing and comparable performance to GPT‑5 and Claude 5. The open‑source community has expressed interest in the model’s architecture, particularly its MoE efficiency techniques.
The launch also raises questions about regulatory oversight and export controls. While the model is open, its large parameter count and multimodal abilities could trigger scrutiny under emerging AI export regulations. No regulatory action has been announced, but industry observers are monitoring the situation.
Looking ahead, Moonshot AI plans to continue expanding the Kimi series. Earlier releases—Kimi 1, Kimi 2, and Kimi 2.5—have progressively increased parameter counts and added multimodal and agentic features. Kimi 3’s release marks the first time an open model has reached the 2.8‑trillion‑parameter scale, a milestone that could influence future open‑weight research and commercial deployments.
In summary, Moonshot AI’s Kimi 3 represents a significant technical achievement and a strategic move toward open AI. The model’s release has already impacted market indices and sparked debate about the future of large language models. Upcoming events include the July 27 weight release, potential partnerships with cloud providers, and continued performance evaluations against U.S. incumbents. The broader implications for AI policy, market competition, and open‑source development remain to be seen.
Founded by drummer Yang Zhilin, a Pink Floyd aficionado, the company’s name—Yuezhi Anmian in Chinese, literally “Dark Side of the Moon”—reflects his musical passion. The international brand, Moonshot, signals an ambition to reach new frontiers. The firm has positioned itself as a challenger to U.S. incumbents such as OpenAI and Anthropic.
Kimi 3 is the first open‑weight model to cross the 2.8‑trillion‑parameter threshold. A post on X announced that the platform’s API and products now grant access to the model, while the full set of weights is slated for release on July 27, 2026. The MoE design activates only a subset of parameters for each inference, improving efficiency for long‑horizon agentic coding tasks. Its multimodal capabilities let it process visual inputs natively, and the 1 million‑token window expands the length of text it can handle in a single prompt.
The announcement places Kimi 3 in direct competition with Anthropic’s Claude 5 and OpenAI’s most advanced closed models. Moonshot AI claims the new model rivals U.S. leaders on several performance benchmarks and notes a pricing structure that undercuts most competitors. The open‑weight nature of Kimi 3 is significant because it allows developers to download, modify, and deploy the model on their own infrastructure—an option unavailable for the current generation of U.S. models.
Market reaction was swift. Within hours of the announcement, the Nasdaq Composite fell, reflecting investor concern over a potential shift in the competitive landscape of large language models. Analysts noted that the release of an open‑weight, 3‑trillion‑parameter model from China could accelerate the spread of advanced AI capabilities beyond the U.S. cloud ecosystem.
Moonshot AI’s strategy to release the full weights a month after the API launch is intended to balance early commercial use with broader community engagement. The company has already positioned Kimi 3 as a professional coding tool, citing lower pricing and comparable performance to GPT‑5 and Claude 5. The open‑source community has expressed interest in the model’s architecture, particularly its MoE efficiency techniques.
The launch also raises questions about regulatory oversight and export controls. While the model is open, its large parameter count and multimodal abilities could trigger scrutiny under emerging AI export regulations. No regulatory action has been announced, but industry observers are monitoring the situation.
Looking ahead, Moonshot AI plans to continue expanding the Kimi series. Earlier releases—Kimi 1, Kimi 2, and Kimi 2.5—have progressively increased parameter counts and added multimodal and agentic features. Kimi 3’s release marks the first time an open model has reached the 2.8‑trillion‑parameter scale, a milestone that could influence future open‑weight research and commercial deployments.
In summary, Moonshot AI’s Kimi 3 represents a significant technical achievement and a strategic move toward open AI. The model’s release has already impacted market indices and sparked debate about the future of large language models. Upcoming events include the July 27 weight release, potential partnerships with cloud providers, and continued performance evaluations against U.S. incumbents. The broader implications for AI policy, market competition, and open‑source development remain to be seen.
