Teh Beng

Hi founders,

I’ve been running the same prompts through Claude, GPT-4o, and Gemini side by side for a few months.

Some patterns I’ve noticed:

→ GPT-4o is confident. Sometimes too confident.

→ Claude tends to caveat more — but those caveats are often the most useful part.

→ Gemini gives a different angle entirely, which can unstick you when the other two are circling the same idea.

The interesting thing isn’t which one “wins.” It’s where they disagree.

When all three agree, I ship it. When two disagree, I read closer. When all three give different answers — that’s where the real thinking starts.

I got tired of managing three tabs and three subscriptions just to do this, so I built ByteChat - bytechat.io - you bring your own API keys, ask all three in one window, and see the answers side by side.

No subscriptions. You pay the model providers directly, usually a fraction of what ChatGPT Plus costs.

Anyone else running multi-model comparisons as a regular workflow? Curious what disagreement patterns you’ve spotted

All comments (or roasting) welcome!