Moonshot hits 65.7% on coding benchmark with Kimi K3.1 Pro

Moonshot has released its flagship artificial intelligence model, Kimi K3.1 Pro, making it available on kimi.ai and through an application programming interface. The software positions Moonshot as a competitive alternative to Anthropic's Claude and OpenAI's latest models for terminal-based agentic coding tasks.

The new model scored 65.7% on Terminal-Bench 4.0, according to published performance metrics. The testing suite is recognized as the leading benchmark for evaluating agentic coding capabilities across terminal environments.

Moonshot committed to releasing open weights for Kimi K3.1 Pro by November 4, 2026, according to company statements. The planned release would provide outside developers with direct access to the underlying model parameters.

Unverified screenshots circulating on social media platforms alleged undisclosed paid influencer promotion surrounding the launch. The claims emerged as the model became publicly accessible for enterprise and developer use.

What do these terms mean?

Agentic coding: An artificial intelligence capability where software agents autonomously plan, write, and test computer code across terminal and developer environments.

Application Programming Interface (API): A software interface that allows external applications to communicate directly with an artificial intelligence model to send requests and receive outputs.

Open weights: The accessible parameters of a trained machine learning model, allowing developers to inspect, host, and run the model on their own infrastructure.