Choose an AI model by testing it on the work you need to do and comparing the result with its cost. The practical way to choose is to group your work by task, try two or three models on a real example, and keep the one whose answers you would actually use. NanthAI Edge is set up for that: it gives you 300+ models in one picker and lets you send the same message to up to three at once.
Models change, and answers depend on the prompt. Use the filters and a comparison on your own task to decide.
What task do you need the model for?
Think about what you will use the answer for. Then group your work into four kinds.
Writing. Emails, drafts, summaries, rewrites and social posts. For this, look at how the model handles tone. Give it a sample of your own writing and ask for a reply in the same voice. A model that writes long and formal may not suit a quick work message.
Coding. Writing functions, explaining an error, reviewing a file. Here the model needs to follow your instructions exactly and show its reasoning. Check the code by running it or reading it closely. Do not trust a model because its answer looks confident.
Research. Questions where you need facts, sources or a summary of a topic. Turn on web search so the model can cite pages. Edge's basic search is free. Pro adds deeper research modes for longer questions. Always open the sources that matter to you.
Images and media. For pictures, you need a model that generates images, not one that only describes them. The Image Gen filter in the picker shows those. Video and music have their own generation features.
How do you narrow the model list?
Tap the + button in the chat input bar and choose Participants. The picker lets you filter by Free, Vision (accepts images), Image Gen and Tools (can call tools such as search). You can sort by 9 criteria, including Coding, Research, Speed, Value and Price, so you can line up the models that fit the task before you read a single answer.
If you are new to this, tap "Help me choose". It takes three steps: you pick your task, then what matters most (quality, speed or value), and it shows the top three models, with scores from independent benchmarks. Treat those scores as a starting list, not a verdict.
How do you test models on your own work?
This is the test that matters. Pick a real task from your week, not a made-up quiz.
- Select two or three models in the picker. Your selection is saved for that chat.
- Send the same message once. Each model's reply appears in one card, stacked under the others.
- Read the replies against the same checklist. Was it right? Did it do what you asked in the length you wanted? Would you have to rewrite it?
- Keep the model that gave the answer you would use, and fork that reply into its own branch to continue.
- If one model was weak, use retry-different to swap just that slot for another model. The rest of the thread stays as it is.
Save the combination you liked as a favourite, so you can start a new chat with the same models in one tap. See Chat Defaults & Favorites for how to set this up.
How do you assess the cost?
Models are priced differently, and a model that gives a slightly better answer can cost far more per message. Turn on Show Advanced Stats in Settings. Each reply then shows its cost, and the chat header shows the running total. In a multi-model chat, each model's cost is listed, with a group total at the bottom. That makes it easy to see whether the extra quality is worth the price for your kind of task.
For everyday messages, a cheaper model is often enough. Keep a premium model for the work where a mistake would cost you more. The Price Transparency page explains the cost numbers in more detail.
When should you test your choice again?
Try again after a few weeks. Models are updated, added and retired, and the one that suited your writing in spring may not suit it now. Rerun your side-by-side test on the same kind of task when you notice the answers drifting.
Keep a short note of what worked for each task, so the choice is quick next time. Your favourite combinations and defaults already keep it in one place.
Try it
Open https://chat.nanthai.tech, pick one task you did this week, and run it through two models side by side. For the picker in detail, see Participant Options. For image models, see Image Generation.
