OpenAI API pricing — every model compared

Last verified

Every OpenAI model with current API pricing, in dollars per million tokens. Verified against OpenAI's own documentation and dated.

All OpenAI models

ModelInputOutputCached inOut:in
GPT-5 nano$0.05$0.40$0.0058.0×
GPT-4.1 nano$0.10$0.40$0.034.0×
GPT-4o mini$0.15$0.60$0.074.0×
GPT-5.6 Luna$0.20$1.20$0.026.0×
GPT-5.4 nano$0.20$1.25$0.026.3×
GPT-5 mini$0.25$2.00$0.038.0×
GPT-4.1 mini$0.40$1.60$0.104.0×
GPT-5.4 mini$0.75$4.50$0.076.0×
o3-mini$1.10$4.40$0.554.0×
o4-mini$1.10$4.40$0.284.0×
GPT-5.1$1.25$10.00$0.138.0×
GPT-5$1.25$10.00$0.138.0×
GPT-5.2$1.75$14.00$0.178.0×
GPT-5.6 Terra$2.00$12.00$0.206.0×
GPT-4.1$2.00$8.00$0.504.0×
o3$2.00$8.00$0.504.0×
GPT-5.4$2.50$15.00$0.256.0×
GPT-4o$2.50$10.00$1.254.0×
GPT-5.6 Sol$5.00$30.00$0.506.0×
GPT-5.5$5.00$30.00$0.506.0×
GPT-5 Pro$15.00$120.008.0×
o3-pro$20.00$80.004.0×
GPT-5.5 Pro$30.00$180.006.0×

The spread

Across OpenAI's current models, input pricing runs from $0.05 (GPT-5 nano) to $30.00 (GPT-5.5 Pro) per million tokens — a 600× range.

That range is the whole reason model selection matters more than prompt golfing. Moving a workload down a tier, where quality allows, changes the bill by a multiple. Shortening a prompt changes it by a percentage.

Reading these numbers

  • Output costs more than input on every model here — 4.0× to 8.0×. A workload that generates a lot is priced very differently from one that reads a lot.
  • Cached input is the biggest discount available. If your prompts share a stable prefix, structuring for cache hits is worth more than any model change short of a tier drop.
  • Token counts are not comparable across providers. Each tokenizes differently, so the same text is a different number of tokens on OpenAI than elsewhere. Cross-provider price comparisons carry a margin of error.

Getting your actual spend

These are list prices. For what you are really being billed, read OpenAI's billing API — an admin key is what that takes.

Source

Verified against OpenAI's pricing documentation. Each model page carries its own verification date.