Bijan Bowen reviews Moonshot’s reported Kimi K3.1 Pro benchmarksA benchmark is a defined test or collection of tasks used to compare a system's performance under specified conditions. and pricing, and notes that the planned open-weight model may have different cybersecurity capabilities from the hosted version. He says he received early access and API credits for testing.
His hands-on tests include a browser operating systemCode generation uses AI or another automated system to create source code from instructions, examples, schemas, or higher-level specifications., a skateboarding game, a robot arm and a wrestling game. He describes improvements after follow-up promptsAn AI feedback loop occurs when AI outputs or their consequences become inputs that influence later model or agent behavior. as well as failed or incomplete tasks, and notes that the robot-arm run used a low reasoning setting and an early version.
Bowen also examines an engine model, browser games and a PC repair game. He reports the cost of several fast-mode testsInference cost is the expense of running a trained AI model to process inputs and produce outputs. and presents his overall judgment as a limited hands-on comparison, not a controlled benchmark.
Watch on YouTube
