Bijan Bowen explains how OpenRouter Fusion combines responses from a panel of models using a judge model, then tests its quality preset on coding and research tasks.
The browser operating system test performs poorly in his assessment, while a skateboarding game and a watch website produce mixed results. He compares a cheaper preset and notes the cost of running multiple models.
Bowen finds the fused response useful for a specific research task about a car, where it highlights leads and blind spots missed in his earlier comparison. He cautions that the small set of tests does not establish general coding superiority.
Watch on YouTube
