Bijan Bowen tests GPT-5.6 Luna at maximum reasoning effort after a major price reduction. The full set of browser, coding, game, three-dimensional and multimodal tasks costs about $1.34, making price and speed the model's clearest strengths.
Bijan Bowen finds the browser-style operating system confusing and the printable engine model poorly adapted to physical constraints. A subway game and C++ city timeline are functional but simple, while a reconstructed television apartment captures the general scene without matching the reference closely.
Bijan Bowen sees better results in a C++ skateboard game, a polished watch site with convincing three-dimensional models and an improved comic game after feedback. The review places Luna as a practical choice for cheap repetitive work that can tolerate supervision, not as a uniformly polished frontier model.
Watch on YouTube
