Learn

Reference

Which AI model for which job?

A quick, honest map of the main models. What each is genuinely best for, how fast it feels, how it lands on cost, and whether there's a free tier. No per-token prices, no benchmark willy-waving: those go out of date the week you read them.

Last reviewed July 2026bands, not prices

Model comparison table

Last reviewed: July 2026Models change fast. Treat this as a signpost, not gospel, and re-check the maker's own page before you commit.
Model / familyBest forSpeedCostReads at onceFree tierNotesAs of
Claude Opus 4.8AnthropicClosedThe hardest thinking: long documents, careful analysis, code.Considered£££Very large. Reads a big pile of documents in one go.No, paid plan or APIThe most capable Claude in the Opus tier. Reach for it when getting it right matters more than speed.July 2026
Claude Sonnet 5AnthropicClosedThe best all-rounder: everyday writing, summarising, most work.Balanced££Large. Handles long chats and sizeable documents comfortably.Yes, on the free Claude planThe best speed-to-quality balance and a sensible default. Fast enough to feel instant, sharp enough for real work.July 2026
Claude Haiku 4.5AnthropicClosedHigh-volume, quick jobs: tagging, sorting, short replies.Fast£Large, but you mostly use it for shorter tasks.Yes, via the API free allowanceThe fastest and cheapest Claude. Great for automations that run a lot.July 2026
Claude Fable 5AnthropicClosedThe most capable frontier work when nothing else is enough.Considered£££Very large.No, paid accessThe most capable model overall, top of the range. Overkill for most day-to-day business tasks.July 2026
GPT (ChatGPT)OpenAIClosedA strong, familiar all-rounder with a big app ecosystem.Balanced££Large, varies by model tier.Yes, a capable free tierThe best-known option. Well worth trying alongside Claude to see which voice you prefer.July 2026
GeminiGoogleClosedResearch and anything wired into Google Workspace.Balanced££Very large on the top tiers.YesHandy if you already live in Gmail, Docs and Sheets.July 2026
LlamaMeta (open weights)OpenRunning a model on your own kit for privacy or cost control.Balanced£Varies by version.Free to run, you pay for the hostingOpen weights. Powerful, but you or a supplier has to host and maintain it.July 2026
MistralMistral (open weights)OpenEfficient open models for self-hosting and lighter tasks.Fast£Varies by version.Free to run, you pay for the hostingA leaner open option. Same trade-off: you own the running of it.July 2026

Cost is a relative running-cost band (£ cheapest to £££ dearest), not a quoted price. No per-token prices or benchmark scores here on purpose.

How to read it

Pick for the job, not the hype.

  • Start with a strong all-rounder. For most business work, a balanced model like Claude Sonnet 5 or the free tier of ChatGPT or Gemini is more than enough. Reach for the top-end models only when a task genuinely needs the extra care.
  • Match speed and cost to how often it runs. A one-off draft can take the slow, careful model. An automation that fires a hundred times a day wants the fast, cheap one.
  • Open models mean you host it. Llama and Mistral are free to run, but someone has to run and maintain them. Great for privacy or scale, more work to own.
  • Re-check before you commit. These change fast. Treat this as a signpost and confirm the detail on the maker's own page.

Not sure where to start? Themodel picker asks five quick questions and points you at the right kind of model in plain English.

Still not sure which to use?

Bring your actual work to the masterclass and we'll help you pick and set it up on the day. The 13 August session in Manchester is free; £20 a ticket after.

Reserve your seat