Next time someone tells you “such-and-such model is garbage” or “such-and-such is the strongest,” you can tell them: Kids pick one. I’ll take them all.
The point isn’t to swap in a stronger AI. It’s to let Claude be the commanding brain and hand the right task to the right tool. Here’s my actual setup.
Claude: The Central Brain
Claude is the central brain of the whole workflow. Its strongest suits are coding, long-running tasks, and tool calls. Internally it splits the work: Opus, Sonnet, and Haiku each take jobs of different weight.
Gemini: King of Cost-Performance and Multimodal
Long documents and bulk data cleaning go to Gemini, where the cost-performance ratio is best. It’s also the king of multimodal: audio, video, OCR, and layout checks are all within reach. That fills in exactly where Claude is weak: Claude can’t handle audio, video, or multimodal.
GPT / Codex: Backup Brain and Image Generation
GPT / Codex serve as the backup brain, handling second-opinion review and rescue when you’re stuck. They also have the strongest image generation right now (GPT Image 2).
How: Call Other Models Without Switching Windows
I use the Codex Plugin and turn Gemini into an agent, so the primary model can call the others without switching windows. Claude automatically assigns each task to the best-suited model, then pulls the results back to integrate.
The video has two hands-on demos: using Codex to generate IG share cards; and turning Gemini into an agent to do audio transcription plus a 200-page PDF summary.