Moonshot AI launches Kimi K3.1 Pro with better coding skills, reduced cyber risks

Moonshot AI, the Beijing-based artificial intelligence startup, has introduced Kimi K3.1 Pro, its first major model upgrade since K3 gained global attention in July 2026. The new model, which will be announced on Saturday, October 24, is designed to make financial research and software development easier and 3D modelling but restrict its potential application in cyberattacks.

The company terms Kimi K3.1 Pro as a very powerful, open-weight, 3-trillion-parameter AI model. Moonshot will offer its model weights on November 4, which will enable developers to run and change the system without its involvement. The launch comes amid the rising competition and scrutiny of Chinese AI firms over the cybersecurity concerns of more sophisticated AI models.

Kimi K3.1 Pro focuses on coding and financial research

Moonshot claimed Kimi K3.1 Pro is aimed at enhancing coding, financial research and 3D modelling, including game development. The capabilities may enable the startup to compete with top AI builders such as OpenAI and Anthropic.

Kimi K3's successful launch in July sparked worry among investors regarding the cost benefits of Chinese AI development and saw its performance at par with established models. But newer machines from the competition like Z.ai and DeepSeek have since inched up on relevant industry leaderboards.

Moonshot reduces the model's offensive cyber capabilities

Moonshot claimed that it stripped cyber-related data from the data set and reinforcement-learning environments which were used to program Kimi K3.1 Pro. The company sought to not introduce open-weight models; it could see being pushed to the "absolute limits" for offensive cyber capabilities.

The model achieved an 81% rating on its own cybersecurity metric, compared to 90.5% for a similar version that was not built with the risk-reduction features, according to Moonshot. This test was conducted by discovering and exploiting weaknesses in software in a controlled setting. Unindependent results.

Open-weight AI models can be used by the company to analyse and protect their software, including legacy applications, the company said. However, sharing model weights doesn't make it easier to track usage or implement safeguards once the files are out there.

The shift is coming as governments and security agencies have expressed concerns that highly advanced AI systems could allow for the automation of sophisticated hacking activities to become easier and cheaper. Last week, China put forward proposals that would include some AI technologies that are associated with malware and AI tools for discovering vulnerabilities.

The new model offers a million-token context window

With a context window of 1 million tokens, Kimi K3.1 Pro can handle vast amounts of information in one go. Moonshot has also launched Kimi K3.1 Pro Highspeed, which is claimed to run as quick as two to three times faster than the normal version.

Moonshot targets growth amid intense competition

According to Bloomberg, Moonshot recently raised its private funds at about $50 billion valuations, ahead of a possible first-quarter listing in Hong Kong in 2027. Bloomberg additionally noted that the company is aiming for $2 billion in annualised income by the end of 2026, following exceeding $1 billion in annual recurring revenue in August.

The new release reflects Moonshot's strategy to stay in the game of China's rapidly evolving AI landscape and address issues regarding the growing intelligence levels of open-weight models.