Moonshot AI launches Kimi K3.1 Pro with 3-trillion-parameter scale

A major upgrade for Kimi

Moonshot AI, a startup based in Beijing, has introduced Kimi K3.1 Pro as its first significant update since the original K3 model gained attention in July 2026. This new iteration is built to handle complex tasks like software development, financial research, and 3D modeling. It specifically targets industries where processing large amounts of data is a daily requirement.

The model is described as a very powerful, open-weight AI system. By offering it as an open-weight model, Moonshot allows developers to download and run the system on their own hardware. This means the company does not have to be involved in every instance where the model is used or modified by the public.

The launch arrives as Chinese AI firms face increasing scrutiny over the security implications of high-end models. Moonshot is attempting to maintain its position against major competitors like OpenAI and Anthropic. The company aims to prove that Chinese AI can match the performance of established global leaders while navigating new safety requirements.

Navigating a crowded market

The release of Kimi K3.1 Pro comes at a time of intense competition within the global artificial intelligence landscape. While Moonshot AI's previous models were notable, newer machines from rivals like Z.ai and DeepSeek have recently moved up on relevant industry leaderboards. This puts pressure on Moonshot to demonstrate that its latest model can maintain a competitive edge against these rising challengers.

To stay relevant, the company is focusing on specific high-value applications like 3D modeling for game development and complex financial research. By refining these specific capabilities, Moonshot hopes to differentiate its technology from other models that may have similar general-purpose strengths. The goal is to provide a specialized toolset that appeals to professional developers and financial analysts who require high precision.

Addressing global security concerns

The decision to strip cyber-related data from the training environment reflects a broader trend of concern from governments and security agencies. These authorities worry that highly advanced AI systems could make it significantly easier and cheaper to automate sophisticated hacking activities. By intentionally reducing these offensive capabilities, Moonshot is attempting to address the risks associated with the growing intelligence of open-weight models.

This move is particularly relevant given recent policy shifts in China. Last week, the country put forward proposals that would include certain AI technologies associated with malware and tools for discovering software vulnerabilities. Moonshot's efforts to limit these specific functions align with a growing need to manage the dual-use nature of powerful AI, where the same technology used for defense can be repurposed for cyberattacks.

Financial goals and expansion

Moonshot AI is currently pursuing a path of rapid growth to solidify its standing in the industry. According to reports from Bloomberg, the company recently raised private funds at a valuation of about $50 billion. This valuation sets the stage for a potential first-quarter listing in Hong Kong in 2027, marking a major milestone for the Beijing-based startup as it seeks to scale its operations.

The company's financial targets are equally ambitious as it looks toward the end of 2026. Bloomberg noted that Moonshot is aiming for $2 billion in annualized income, following a period where it exceeded $1 billion in annual recurring revenue in August. These figures highlight the company's transition from a startup to a major commercial player, balancing technical innovation with the need for significant financial returns.

The mechanics of open weights

Moonshot AI plans to release the model weights on November 4, which allows developers to run and modify the system independently. This means the company will not be involved in how the system is changed or used once the files are distributed. While this provides freedom for developers, it also creates a challenge for the company because sharing weights makes it harder to track usage or implement safety safeguards once the files are out there.

The company conducted a test to measure these risks by discovering and exploiting weaknesses in software within a controlled setting. This specific testing method allowed them to evaluate how the model behaves when pushed toward offensive cyber capabilities. By identifying these vulnerabilities, Moonshot can better understand the limitations of the model they are releasing to the public.