OpenAI API Pricing (2026)

OpenAI’s 2026 lineup runs from GPT-5.4-nano at $0.20 per million input tokens up to GPT-5.5 at $5. Cached input costs 10% of the standard rate and the Batch API takes 50% off both input and output. Pick nano for cheap classification, GPT-5.4 as the general workhorse, and GPT-5.5 only for the hardest tasks. Rates below are updated 2026 from OpenAI’s pricing page and will keep moving.

Per-million-token rates (updated 2026)

Model	Input	Cached input	Output	Use it when
GPT-5.5	$5.00	$0.50	$30.00	Hardest reasoning and agent tasks
GPT-5.4	$2.50	$0.25	$15.00	General-purpose workhorse
GPT-5.4-mini	$0.75	$0.075	$4.50	Good quality at a fraction of the cost
GPT-5.4-nano	$0.20	$0.02	$1.25	High-volume, simple tasks

Long-context variants of the larger models are billed at higher rates. The o-series reasoning models and GPT-4.1 variants are listed separately on OpenAI’s full pricing page.

Two discounts that matter

• Cached input costs 10% of the standard input price. If you resend the same system prompt or document on every call, caching pays for itself fast.
• Batch API gives a 50% discount on input and output for asynchronous work. GPT-5.4 batch input drops to about $1.25 / 1M.

Note: regional data-residency endpoints carry a 10% uplift for models released on or after March 5, 2026.

Honest read

OpenAI is rarely the outright cheapest on raw input price. Gemini 2.5 Flash-Lite and Mistral Small undercut nano. Where OpenAI competes is ecosystem maturity and the mini/nano tiers being genuinely capable for routing. If price is the only axis, compare against Gemini and Anthropic before committing.

Related

See the full cheapest LLM API comparison for the cross-provider table, and cost control for how to actually cut the bill. Background: tokens and context windows. For help wiring OpenAI into a product, see AI integration.