Whatever you have access to already. Thanks for reading.
But no, seriously: if you don’t have a strong opinion, the models are all getting so good so fast that if one gains an advantage over the others, the rest will have it baked into their next release a few months later.
Claude
ChatGPT
Gemini
DeepSeek
Qwen
Llama
Traditionally Claude has had a bit of an edge for coding, and for design taste when making slide decks and infographics. OpenAI’s built-in image generation is better, and Gemini is fast and speedy. But this changes every few months.
| Where it shines right now | Watch out for | |
|---|---|---|
| Claude (Anthropic) | Coding with Claude Code, long documents, design taste in slides and infographics | Usage limits on busy days |
| ChatGPT (OpenAI) | Built-in image generation, Codex, the biggest ecosystem | Usage limits, frequent model renames |
| Gemini (Google) | Speed, price, huge context, Google Workspace | Quality varies by task |
| Open source | Cheap, and you can run it yourself | Lags a few months; data sovereignty (below) |
I was using Codex and OpenAI models for most of my coding until a few weeks ago, when the usage limits started hitting me hard. So I switched to Claude Code from Anthropic, and I’m now happily moving along there. That will switch again in a few months’ time, and again a bit later, as new releases come out.
But they all write great code, can operate my Mac really well, and turn my voice notes at the gym into ready-to-present slide decks by the time I get home.
The differences are marginal. Rate limits, cost, how easy the UI is and switching costs increasingly matter more.
The one strong opinion: Chinese open models
While open-source models generally lag OpenAI and Anthropic by a few months, they’re often cheaper and will get your work done just as well. But I’d caution you about data sovereignty before sending all your data East, no matter what the policies say. DeepSeek’s own app, for example, stores what you send on servers in China, where companies can be compelled to hand data to the state.
Self-hosting open models is possible and definitely works (and keeps your data at home), but it will be prohibitively expensive compared to the OpenAI and Anthropic APIs. Be prepared to pay ten times as much and get less reliability.
Those services have zero-data-retention options and enterprise tiers anyway, which I’d strongly encourage you to explore first. Invest in anonymisation and personal-information redaction on your own machines before using online models at all. A hybrid system is often best:
- Redact names, ID numbers and account details locally first.
- Send the cleaned work to an enterprise Claude or ChatGPT account with zero retention.
- Keep the truly sensitive stuff on a small model running on your own hardware.
The short version
Use whatever you’ve got already. Sign up for a team Claude or ChatGPT account and start prompting. The products and models are all converging anyway.