Kimi K3.1 Pro is 4th on the Artificial Analysis index with 55 points

Moonshot AI made a new model, Kimi K3.1 Pro, and the price of the model is 33% lower. It is 4th on an index of 10 tests for code, science and agent work (work in many steps), with 55 points. Only 3 models have more points (56 to 58), and the next 2 models of other companies have 53 points.

What is new

Artificial Analysis made the index. The model has 11 more points than Kimi K3. The increase is most in agent work. On Terminal-Bench 4.0, a test of code in a terminal, it has 67.2%, the top result of all models. Kimi K3 had 13%.

The cost

The cost of a task is about the same as with Kimi K3. The price is 33% lower, and the model uses 14% more output tokens.

What is not known

Moonshot has no open weights (the model files that a person can get) or license at this time. Moonshot reports that it plans to make them available in the future.

Insight: A high rank on one index does not show how a model does on your task

In a 3D test from Artificial Analysis, Kimi K3 is 11th, more than 200 points below the first model. The report on K3.1 Pro gives no 3D result. Moonshot reads as a maker that works on agent tasks and low cost, and open weights can push the price of closed models down. Test the model on your tasks.