What's the difference between input and output token pricing?

Input tokens are the words you send to the API (your prompt), while output tokens are what the model generates in response. OpenAI charges separately for each, with output tokens typically costing 2-4x more. For example, GPT-4o charges $2.50 per million input tokens but $10.00 per million output tokens.

Does OpenAI charge for API requests that fail or error out?

No, OpenAI only charges for successfully processed tokens. If your API request fails due to server errors, rate limits, or invalid parameters before token processing begins, you are not charged. However, if the model starts generating a response and you cancel it mid-stream, you pay for tokens generated up to that point.

How does OpenAI API pricing compare to Anthropic Claude?

Anthropic's Claude 3.5 Sonnet costs $3.00/$15.00 per million tokens (input/output), making it 20% more expensive than GPT-4o at the input level but 50% pricier for outputs. Claude 3.5 Haiku at $0.80/$4.00 is significantly more expensive than GPT-4o mini's $0.15/$0.60 for budget use cases.

Are there volume discounts for high-usage OpenAI API customers?

Yes, OpenAI offers graduated volume discounts that automatically apply once you exceed certain monthly token thresholds. Enterprise customers processing billions of tokens can negotiate custom pricing. However, standard pay-as-you-go users at lower volumes pay the published list prices without discounts, and OpenAI does not publicly disclose the discount tier thresholds.

PROVIDER GUIDE · VERIFIED 13 JUL 2026

How much does OpenAI's GPT API cost?

OpenAI's GPT API pricing ranges from $0.15 per million input tokens for GPT-4o mini to $10.00 per million for GPT-5.5, with GPT-4o priced at $2.50/$10.00 (input/output). All models charge separately for input and output tokens, with output typically costing 2-4x more than input, making your final cost heavily dependent on response length.

OpenAI's lineup spans the widest price range of any provider we track — a 33x gap between its cheapest and most expensive output token. Here's what that spread actually buys you.

Model	Input /1M	Output /1M	Context
GPT-4o mini	$0.15	$0.60	128K
GPT-4o	$2.50	$10.00	128K
GPT-5.5	$5.00	$30.00	400K

GPT-4o mini is doing more work than its name suggests

At $0.15/$0.60 per million tokens, GPT-4o mini is one of the two cheapest models we track across every provider — tied with Google's Gemini 2.5 Flash. For high-volume tasks where the "mini" label makes it feel like a compromise, remember: it's priced for exactly the classification, extraction, and routing work that makes up most of a typical production system's actual call volume. Defaulting everything to GPT-5.5 out of habit is one of the most common ways teams overspend on OpenAI's API.

GPT-4o's awkward middle position

GPT-4o sits at $2.50/$10.00 — priced close to Anthropic's Claude Sonnet 5 ($2.00/$10.00) and noticeably above Google's Gemini 3 Flash ($0.50/$3.00) for what's often comparable mid-tier capability. If your workload can tolerate a bit less polish for real cost savings, it's worth benchmarking Gemini 3 Flash against GPT-4o for your specific task before defaulting to OpenAI here.

GPT-5.5 is expensive, but not the priciest we track

At $30.00 per million output tokens, GPT-5.5 sits near the top of the flagship tier — more expensive than Anthropic's Claude Opus 4.8 ($25.00) and Google's Gemini 3.1 Pro ($12.00), though still below Anthropic's most capable tier, Claude Fable 5, at $50.00. Reserve GPT-5.5 for tasks where its specific reasoning strengths matter, not as a default "best model" choice — check the full price table before assuming the newest flagship is automatically the right (or the most expensive) pick for your task.

How we'd actually pick

High-volume, simple tasks: GPT-4o mini — but benchmark against Gemini 2.5 Flash first, they're priced identically.
General production workload: GPT-4o — check whether Gemini 3 Flash handles your task at a fraction of the cost before committing.
Maximum capability, cost secondary: GPT-5.5.

Prices verified against platform.openai.com/docs/pricing, 13 July 2026. See full methodology. Use the calculator to compare GPT against every other tracked provider for your own usage.