OpenAI API pricing in TokenCalculator covers GPT-5, GPT-4.1, GPT-4o, Mini, Nano, and o-series rows. Rates checked 12 Aug 2026. Exact in-browser counts use OpenAI encodings.
OpenAI rates in TokenCalculator
Example cost uses 1000 input tokens and 500 output tokens. Rates checked 12 Aug 2026. Open a row to price your own prompt.
OpenAI families
Treat the catalog as a map of product tiers, not a single price. Mini, nano, flash, and flagship rows exist so you can match quality to budget instead of paying frontier rates for classification.
- Nano and Mini: high volume chat and classification.
- GPT-4o and GPT-4.1: general production defaults.
- GPT-5: frontier quality when retries on a smaller model would cost more.
- o-series: reasoning workloads billed as their own rows.
Snapshot
| # | Model | Input / 1M | Output / 1M | Example |
|---|---|---|---|---|
| 1 | GPT-5 Nano | $0.050/1M | $0.400/1M | $0.00025 |
| 2 | GPT-4.1 Nano | $0.100/1M | $0.400/1M | $0.0003 |
| 3 | GPT-4o mini | $0.150/1M | $0.600/1M | $0.00045 |
| 4 | GPT-5.6 Luna | $0.200/1M | $1.20/1M | $0.0008 |
| 5 | GPT-5 Mini | $0.250/1M | $2.00/1M | $0.00125 |
| 6 | GPT-4.1 Mini | $0.400/1M | $1.60/1M | $0.0012 |
| 7 | GPT-3.5 Turbo | $0.500/1M | $1.50/1M | $0.00125 |
| 8 | o1-mini | $1.10/1M | $4.40/1M | $0.0033 |
Levers that change the bill
- Cached input when the prefix is stable and the catalog lists a cache rate.
- Batch API at lower rates when latency can wait.
- Output length, which dominates GPT-5 class bills.
How to estimate in TokenCalculator
Pick the OpenAI row that matches the API model id you deploy. Paste a real prompt, set expected output tokens, then scale by users and messages per day. Toggle cache or batch only when this catalog publishes those rates for that row.
Common mistakes
- Budgeting from words instead of Exact tokens.
- Picking a marketing name that is not the API model id you call.
- Ignoring cache write or batch eligibility.
Frequently asked questions
Is this the official OpenAI price list?
No. It is a curated snapshot with a checked date and a link to OpenAI’s pricing page. Use the official page for contracts.
Why is my invoice different?
Token counts, output length, cache writes, batch jobs, and private discounts all move the invoice. Hold the prompt constant in the calculator, then compare to billed tokens from the provider export.