Bijan Bowen tests Mimo V2.6 Pro and Flash across games, a vehicle showroom and a changing city scene. After tool-call problems across several coding toolsAn AI coding agent is a tool-using AI system that can inspect, modify, and validate software within a repository., Bijan Bowen switches to the native harnessAn AI agent harness is the software framework that packages a model with tools, instructions, context management, execution controls, and user interaction. on Windows but still encounters loops. The review presents a mixed collection of experiments, not a controlled benchmarkA benchmark is a defined test or collection of tasks used to compare a system's performance under specified conditions..
Bijan Bowen finds Flash particularly effective at implementing a detailed racing-game design and building a city that changes across time periods. Flash also improves a guitar-store game and revises a skateboarding prototype, although geometry, controls and moving labels remain imperfectAgent completion verification checks evidence that an agent achieved the requested outcome rather than merely producing intermediate work.. Pro produces a pleasing role-playing recreation after a follow-up prompt and improves a subway game when given concrete feedback.
Bijan Bowen explicitly acknowledges accidentally combining prompts, selecting the wrong prompt and removing a file that the agent was trying to read. Those confounds weaken direct model comparisons, while repeated tool loopsAn agent loop repeatedly observes a task's state, chooses an action, checks the result and decides whether to continue or stop. remain a practical concern. Bijan Bowen prefers Flash in these examples without establishing that Pro is generally inferior.
Watch on YouTube
