Bijan Bowen introduces Kimi K2.7 Code and discusses the published model specifications and claimed reduction in thinking-token useAI token efficiency measures how effectively a model or workflow turns consumed input and output tokens into useful results.. He tests an interactive browser desktop, a subway scene and game, and several C++ projectsCode generation uses AI or another automated system to create source code from instructions, examples, schemas, or higher-level specifications..
Bowen finds the subway and skateboarding examples improved relative to his earlier tests, but observes repeated editing errors, stalled quality checks and unfinished simulations. An image-guided laptop reconstruction captures some interactive features while getting substantial geometry wrong.
A watch website and a 3D printer simulation show stronger individual capabilities, though the printer rendering is flawed. Bowen treats the tests as an informal comparisonEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. and notes that testing immediately after a stronger model may have affected his expectations.
Watch on YouTube
