OpenAI vs Anthropic vs Google: API Pricing Compared

OpenAI, Anthropic and Google cover most production LLM traffic. Their pricing philosophies differ enough that the "cheapest" answer depends entirely on your workload.

The headline numbers

• OpenAI — GPT-5.4 at $2.50 in / $15 out; GPT-5.4 nano at $0.20 / $1.25. A clean flagship-plus-cheap-tier split across the GPT-5 line.
• Anthropic — Claude Sonnet 4.6 at $3 / $15; Haiku 4.5 at $1 / $5. Pricier sticker, but the best-in-class prompt caching (cached input as low as $0.30/1M) often makes it the cheapest in practice for repeated context.
• Google — Gemini 2.0 Flash at $0.10 / $0.40 and 1.5 Flash-8B cheaper still. The cheapest capable models on the market, with a 1M-token context on the Gemini 2.5 line.

Who wins for what

• Cheapest at volume: Google Gemini Flash / Flash-8B, by a wide margin.
• Repeated long context (RAG, agents): Anthropic, once caching is in play.
• Ecosystem & tooling: OpenAI, with the broadest SDK and integration support.

The caching asterisk

Sticker prices hide the biggest variable. If 80% of your input is the same document every call, Anthropic's caching can make a $3/1M model behave like a $0.60/1M one — beating a cheaper provider that has no caching. Always model your real input/output split rather than comparing flagship list prices head-to-head.

See live numbers on the OpenAI, Anthropic and Google Gemini pricing pages, or compare all three in the calculator.