GPT-6 Luna First Test – Is OpenAI’s CHEAPEST Model Actually Good?

In creator Bijan Bowen's narrated first test of OpenAI's GPT-6 Luna, one task shows what a single follow-up prompt did to a generated game. Bowen asked for a 3D subway first-person shooter, and he describes the first version as quite dark and quite simple, with weapons that were tough to see. He then sent one follow-up asking for a full visual overhaul: brighter, better models for everything (the enemies in particular), bullet holes in the environment, and fine detail. By his account the overhaul took about 30 minutes. He reports a bit more visibility, bullet holes that he thought were implemented nicely, and enemies that were significantly improved and no longer just simple shapes. This is one creator's observation of one prototype, and the footage was not reviewed here, so it shows what he saw in that session and not how reliably the model handles revision requests.