OpenAI API Pricing (October 2026)

The headline rates apply to short-context requests. On the GPT-6 tiers, a prompt above 272K input tokens bills the whole request at 2x input and cache rates and 1.5x output: GPT-6 Astra rises from $10/$50 to $20/$75 per million input/output tokens, GPT-6.1 Sol from $2/$10 to $4/$15, and GPT-6 Luna from $0.10/$0.50 to $0.20/$0.75.

GPT-5.6 Sol has its own long-context tier ($8/$30 above 272K input tokens), and GPT-5.4 also lists higher long-context rates. The calculator above uses short-context rates, so long prompts need the higher meter applied before a budget is approved. OpenAI also lists a 10% regional-processing uplift where regional processing is available.

Cached input is automatic: repeated prompt prefixes (system prompts, tool definitions, shared context) bill at the cached rate, $1 per million instead of $10 on GPT-6 Astra. Cache writes on the GPT-6 tiers cost 1.25x standard input, so a prefix that is read back even once costs less than sending it twice uncached. GPT-6.1 Sol's cached rate is 5% of its input price, half GPT-6 Sol's.

The Batch API takes 50% off input and output for asynchronous jobs. On the GPT-6 tiers, Flex is also billed at half the standard rate and Fast mode at 2x. Model your own mix with the LLM API pricing calculator or the estimator above.

An OpenAI API key has no monthly subscription fee. Usage is billed from the token, tool, storage, and processing meters attached to the API account. ChatGPT Free, Go, Plus, Pro, Business, and Enterprise are separate products with per-user or plan pricing; paying for ChatGPT does not create a pool of API credits.

Use this page for application and agent workloads billed through an API key. For the consumer-plan decision, see the current ChatGPT Plus analysis. For hosted OpenAI models with Azure routing and residency controls, compare the Azure LLM pricing table.

Start cheap and move up when your own evals show a quality gap. GPT-6 Luna ($0.10/$0.50) covers classification, extraction, and routing. GPT-6.1 Sol ($2/$10) costs 20x Luna on both rates and is the middle tier. GPT-6 Astra ($10/$50) is the flagship at 5x Sol, for the coding, reasoning, and agentic work where Sol falls short in testing. GPT-5.5 Pro ($30/$180) costs about 3x as much as Astra on input and 3.6x as much as Astra on output, so it only makes sense when evaluation results show Astra isn't enough.

The older tiers no longer save money: GPT-5.6 Sol's promotional $4/$20 and GPT-5.4's $2.50/$15.00 both cost more than GPT-6.1 Sol. Staying on them is a choice about behavior you've already validated, not about price.

GPT-6 Astra lists $10/$50 per million input/output tokens and GPT-6.1 Sol lists $2/$10. Claude Fable 5.1 lists the same rates as Astra. Claude Sonnet 5 matches Sol. Gemini 3.1 Pro lists $2/$12 on prompts up to 200K tokens, the same input rate and about 20% more than Sol on output. GPT-5.6 Sol's promotional $4/$20 matches Claude Opus 5.5.

With prices this close, benchmarks decide it: see the table above next to the Claude API pricing hub, or pick any pair on the compare hub. GPT-6 Astra compared with Google's Gemini 3.8 Flash puts the benchmark scores beside the rates.