I’ve Followed Model Rankings While Building a Side Business

Last Wednesday night, I was helping a freelance photographer in Hangzhou rewrite replies for a client. Halfway through, the old model changed its tone. In a rush, I switched to the newly ranked #1, Gemini 4 Neon. I spent 40 minutes switching back and forth, and the copy didn’t improve. I just started checking every formatting detail again. That was when I realized a ranking can feel like a promotional text arriving before closing time: worth reading, but not worth paying for on the spot.

What the Ranking Is—and Who’s Using It

The Models Engineering Index is a public model-evaluation page maintained by Aarvin Ghosh. It combines different tests and currently places Gemini 4 Neon at #1. It’s more useful for developers tracking model capabilities than as a shopping list for “best model for client work.”

I know an independent-store owner, Lin Yue, who runs a business in Shanghai and bulk-generates product descriptions for three clients. She didn’t switch because of the ranking. Instead, she tested a small batch of the same products, comparing output quality, revision rounds, and actual time spent. She even tracked separately whether the results could be smoothly rewritten into her established voice—a step I used to ignore, and one I’ve definitely handled badly before.

What It Costs to Replicate This Today

Money: standard chat access is usually free to start; check the account page for paid-plan requirements. Time: about 30–60 minutes. Technical barrier: no coding required, but you need to prepare realistic tasks and compare them fairly. First step: open the Gemini 4 Neon page linked from the ranking or its relevant product entry point, find the “Try” or “Get Started” button, and run a test set using client copy without changing the default settings. Then run the same inputs through your current model and record how many revisions each one needs.

There’s no urgency to test today: rankings change quickly, older models don’t necessarily stop working overnight, and this tool isn’t essential for everyone.

Depending on Where I’m at

If I’ve just started, I’d choose one primary model and put the prompt requirements into a reusable template rather than chase the rankings. If I have one or two stable clients, I’d compare the model against my latest three real deliverables, focusing on revisions and consistency across consecutive outputs. If I’m scaling, I’d watch pricing, speed, data-use rules, and how difficult it is to replace the model, so one tool doesn’t lock my team’s workflow in place.