How to choose the right AI model, whether you use Claude, ChatGPT, or Mistral

Every provider names its models differently, and those names change every few months. These changes are a source of confusion (at least it is for me).

Under the names, all major providers follow the same pattern: a fast and inexpensive tier, a balanced everyday tier, and a most capable tier for complex work. Once you see the pattern, choosing the right model for a task becomes much simpler.

This guide gives NGO and charity teams a good way to choose, so you are not relearning the landscape every time a new model launches.

The pattern that does not change

Remember the row, not the names above it. The specific names will keep changing, the three-tier structure will not.

If you use Claude

Claude has the most stable naming of the three: a small tier, a mid tier, and a top tier, each receiving periodic version updates.

• Haiku: quick answers, simple drafts, and high-volume tasks where speed and cost matter more than depth
• Sonnet: the default for most everyday work, including writing, analysis, and coding
• Opus: complex reasoning, high-stakes documents, and problems that need genuine depth rather than speed

For almost everything you do day to day, Sonnet is the right starting point. Reach for Opus when a task specifically needs more careful thought, not as a default.

Personally, I have been using Claude since it was launched. Not for a professional reason. My mother-in-law is called Claude, so I am well used to someone having the answer to everything and hallucinating regularly 😉

I haven't mentioned Anthropic Fable 5 or 5.1 because I don't recommend them for most NGOs. The problem is what happens to your data. Both keep a copy of everything you type for 30 days, for what Anthropic calls “safety and monitoring”. Normally you can ask a provider not to keep your data at all. This is called a Zero Data Retention (ZDR) agreement. Since September 2026, large enterprise customers can apply for a temporary exemption, but on consumer plans (Free, Pro, and Max) retention still applies, so anything private you enter could sit on Anthropic's servers for a month. Microsoft reportedly removed Fable 5 from its internal tools for this very reason. For NGOs that handle donor, beneficiary, or safeguarding data, the same caution applies until every customer can turn data storage off.

If you use ChatGPT

OpenAI renames and re-tiers its lineup more often than the other two providers, sometimes several times within a single year. Rather than memorising a name that may already be outdated, look for the same three-tier pattern in the model picker: a fast option, a balanced default, and a most capable option that you usually have to select deliberately.

As of September 2026, the ChatGPT picker shows thinking levels rather than model names. On Free and Go, you get one model and a Think button for harder questions. On Plus, you choose between Instant, Medium, and High. Pro and Business plans add Extra High and Pro, and Pro gives access to GPT-6, OpenAI's most capable model. The model names themselves (Luna, Terra, Sol) only appear in ChatGPT Work, Codex, and the API. Whatever the current names are when you read this, that structure is what to look for.

I stopped using ChatGPT for ethical reasons. Have a look at quitgpt.org.

If you use Mistral

Mistral keeps the same Small, Medium, and Large naming across generations, which makes it the easiest of the three to track over time.

• Small: fast and inexpensive, strong at classification, summarising, and structured extraction, and available as an open-weight model you can self-host
• Medium: the balanced tier and a sensible default for general work; the latest version is strong enough that the gap with Large has narrowed
• Large: the flagship, strongest at complex reasoning and multilingual work, and particularly competitive in European languages

Although I don’t find it quite as capable as Claude, Mistral is a European model and a more ethical choice. Another ethical option is Euria, the AI assistant from Infomaniak. It is hosted entirely in Switzerland, built on leading open-source models, and designed so your data stays yours: it is never stored or used to train the AI. It also runs on renewable energy and meets both Swiss (LPD) and EU (GDPR) data rules, so it is well worth a look if you handle sensitive information.

A quick decision guide, whatever you use

• A quick question or a simple draft: use the fast tier
• Everyday writing, analysis, or coding: use the balanced tier
• Complex reasoning, a high-stakes document, or anything that genuinely needs depth: use the most capable tier

One habit worth keeping

Within any given tier, always use the newest version available. Older versions often stay available for a while (OpenAI usually retires them around 90 days after a successor arrives), so you may see two or three versions of the same tier side by side. There is almost never a reason to choose an older one when a newer version of the same tier exists.

Why this matters for NGOs

For mission-driven organisations, the choice is not only about quality. It is also about cost, data protection, environmental impact, and where your data lives.

Match the tier to the demands of the task, not to the newest headline. Most day-to-day work belongs in the balanced tier, with the most capable tier reserved for genuinely complex or high-stakes work.

Need help choosing and using AI safely?

On this page, I can’t simply tell you: choose this or that model. It depends on your values, your use cases, your budget, and your level of collaboration. This was meant as a quick guide. You already know that I help charities and not-for-profit organisations turn scattered AI experimentation into a practical, responsible operating model. and here comes the sales pitch

A short session can help you:

• Match the right tools and tiers to your real tasks and budget
• Clarify data red lines and residency requirements
• Set simple governance and human review rules
• Build staff confidence with examples from their own work