Which AI platform is cheapest for companies? Claude vs ChatGPT vs Gemini

The short answer

If you only look at price per token, Google's Gemini models are usually the cheapest in the middle and top tiers, and the three providers are almost level on their smallest models.

But the biggest saving doesn't come from switching provider. It comes from using a small, cheap model for simple jobs and keeping the expensive models for work that needs them. That can cut a bill by 10 to 100 times on the same platform, as the task table below shows.

Model prices side by side

Each provider sells a family of models, from small and fast to large and most capable. Here's one model from each tier.

Swipe sideways to see every provider

List prices in US dollars per million tokens for the standard API, checked on 2 November 2026. OpenAI from its pricing page; Claude and Gemini from a pricing tracker that checks the official pages. [Verify each against the provider's own pricing page before publishing.] Prices change often.

Claude also has Opus 5.5 ($4 in / $20 out), which sits between Sonnet and Fable and is often enough for demanding work.

What common tasks cost on each

Token prices are hard to picture, so here are five everyday jobs. Each figure is the cost of doing that task 1,000 times on the tier we'd normally use for it, at the list prices above.

Swipe sideways to see every provider

Worked out from the list prices above, before caching or batch discounts. Real token counts vary with your prompts and documents.

Two things stand out. Simple jobs cost pennies on any platform. And the same customer reply costs $5.50 per thousand on a balanced model but around $0.28 on a budget model. If the budget model's replies are good enough, that choice saves more than any switch between providers.

Why the cheapest price isn't always the cheapest bill

Quality. A cheaper model that gets one in ten answers wrong costs more in staff time than it saves. Always test each model on your own examples before choosing.

Caching and batch. All three providers offer prompt caching and batch processing, and both lower the effective price a lot. A tool built well on a dearer model can cost less than one built badly on a cheap model.

Price changes. Providers change prices and launch new models every few months, and some prices are promotional. Build so that you can swap models without rebuilding everything.

Seats vs API. If your team mainly uses ChatGPT, Claude or Gemini through the apps, you pay per person, not per token. The comparison above applies to automations and tools that use the APIs.

For a closer look at each provider, read our guides to Claude API pricing and OpenAI API costs, and what to do if you're hit with a surprise Gemini API bill. Our main guide to AI bills explains where the waste usually hides.

How to find the cheapest option for your business

1. 01List the jobs. Write down each task you want AI to do and roughly how often.
2. 02Start small. Try each job on the budget tier first and only move up if quality isn't good enough.
3. 03Test on real examples. Score the results with the people who do the work.
4. 04Add caching and batching where the same instructions repeat or the work can wait.
5. 05Keep it portable. Use a set-up that lets you switch models when prices change.
6. 06Review every quarter. New models often do the same job for less.