AICodeKing evaluates Kimi K3 through eight coding, animation and reasoning tasks, describing sponsor-funded early access while saying the sponsor did not control the review. The model earns 62 of 80 points in the creator's local benchmark, with strong elevator and folding-table results but weaker contact-lens and wristwatch interactions.
A longer autonomous task combines dataset creation, model fine-tuning and a local web interface. The creator highlights recovery from errors and browser-based checks, while distinguishing personal workflow impressions from single-shot benchmark scores. Whether the model will be openly released is explicitly unconfirmed in the video.
Watch on YouTube
