My cheat sheet, as of early November 2026. It's how I use them, not a benchmark. Prices and plans change often: check the app before you pay.
#The one-page table
| Job | My pick | Why |
|---|---|---|
| Long documents, careful writing | Claude | reads long PDFs well, admits when it's guessing |
| Doing things on your computer | ChatGPT | its newest model (GPT-6.1 Terra) can look at the screen, click and type, in ChatGPT and Codex |
| Your Google stuff (Gmail, Drive, Docs) | Gemini | it's already inside your Google account |
| Photos of things (fridge, wardrobe, receipts) | any of the three | all read photos; ask them to list what they see first |
| Private, stays in Europe | Le Chat by Mistral (Paris) | European company, open models you can also run yourself |
| Coding | Claude Code or Codex | both work in your terminal; use tests either way |
#"Which one is the smartest?"
On the Artificial Analysis ranking (late October 2026), Google's new Gemini 4 Carbon scored 59, one point above Claude Opus 5.5. You can't use Carbon yet: Google is giving it to selected users first. For everyday tasks the difference between the top models is smaller than the difference a good prompt makes.
#What the free plans give you
All three have a free plan with a daily limit, photo upload included. The paid plans (around 20 a month) mainly give you more usage, the newest models first, and agent features. Check the exact limits in each app, they change.
#The one habit that matters more than the model
Ask for the source. Every model sometimes states made-up numbers confidently. See sangeeths.ai/zombie.
#Sources
- GPT-6.1 Terra: OpenAI
- Gemini 4 Carbon: 9to5Google, 28 Oct 2026
- Mistral Large 4: Mistral
Made by @sangeeths.ai: real AI engineering, tested live. Share this kit: sangeeths.ai/models · All free kits: sangeeths.ai