Wes Roth compares reported coding benchmarks with early Grok 4.5 experiments rather than presenting a controlled ranking. A sailing prototype responds to wind and weather, adds capsizing behavior after feedback and combines generated geometry with external synthetic audio. The demonstration still contains clipping and effects that need refinement.
Wes Roth's role-playing and smaller 3D experiments show interactive dialogue, branching choices and playable scenes, but also pointer problems, clipping, inverted roofs and lighting issues that require follow-up. These are narrow prototype checks, not evidence of reliable production gameplay or complete model superiority.
Wes Roth then uses a stronger planning model to specify a large city and sends individual district tasks to Grok 4.5. The point is to separate architectural planning, bounded implementation and output checking instead of asking one model to do everything. Reported costs and launch claims remain his account of a small early test; future business performance, training recipes and unreleased models are speculative.
Watch on YouTube
