Full price table — your workload
Reference $/1M tokens (July 2026), monthly cost on your numbers above.
| Provider | Model | In $/1M | Out $/1M | Cost / mo |
|---|
Cheapest by use case
- High-volume / simple (classification, extraction, routing): Gemini 2.0 Flash or GPT-4o mini — pennies per thousand requests.
- Chatbots & support: GPT-4o mini or Gemini 2.5 Flash — cheap, fast, good enough; see the chatbot cost calculator.
- RAG / long context: Gemini 2.5 Flash or Pro (low input price helps when you stuff in retrieved chunks); see the RAG cost calculator.
- Hard reasoning / coding: Claude Sonnet 4 or GPT-4o; Claude Opus / o3 only when you truly need it.
- Need a free tier: Gemini (AI Studio) is the only one of the three with a genuine free quota.
How to actually choose
Brand loyalty is the most expensive habit in AI. The three providers leapfrog each other constantly, and the real cost difference comes from two things you control: the model tier you pick and how long your outputs are. A frontier model on short answers can be cheaper than a "cheap" model on rambling ones. Price the workload, not the logo — the table above does exactly that on your own numbers.
Drill into one provider with the GPT-4o calculator, the Claude calculator or the Gemini calculator, compare all models at once on the full AI API cost calculator, or estimate a whole product with the AI app cost estimator.
Three levers cut the bill further once you've picked a model, and all three providers offer all three. Prompt caching reprices repeated system prompts and documents at a fraction of the normal input rate — worth it for any app that resends the same instructions or context on every call; size it on the prompt caching savings calculator. A batch API gives roughly 50% off for jobs that can wait a few hours instead of needing a live response; see the batch API savings calculator. And for high-volume apps, routing routine requests to a cheaper model while reserving the flagship for hard cases is often the single biggest lever of all — model it on the AI model router cost calculator.
One more difference that doesn't show up in the price table: only Gemini has a genuine free API tier (Google AI Studio, no card required), which makes it the cheapest way to prototype before committing spend. OpenAI and Anthropic both require billing set up from the first request, though new accounts get limited trial credit.
FAQ
Which is cheapest: OpenAI, Claude or Gemini? For the very cheapest capable model, Gemini 2.0 Flash ($0.10 / $0.40 per 1M tokens) and GPT-4o mini ($0.15 / $0.60) lead. Among flagships, Gemini 2.5 Pro has the lowest input price, GPT-4o sits in the middle, and Claude Opus is the most expensive — the winner depends on the tier you need, not the brand.
Is Claude more expensive than GPT-4o? Claude Sonnet 4 ($3 / $15) is a little pricier than GPT-4o ($2.50 / $10), and Claude Opus 4 is far more expensive — but Claude Haiku 3.5 undercuts GPT-4o. Compare on the workload you actually run, since output length dominates the bill.
Do all three support prompt caching and batch discounts? Yes — all three offer caching for repeated prompts and a batch API at roughly half price. The real savings depend on how repetitive your prompts are and whether your job can tolerate the batch delay.
Does any of the three have a free API tier? Gemini is the only one — Google AI Studio gives free access with rate limits generous enough for prototyping. OpenAI and Anthropic require billing set up first, though both give limited trial credit to new accounts.