How OpenAI API pricing is billed
OpenAI charges separately for tokens the model reads (input) and tokens it writes (output). Rates are published per million tokens. A request that sends 2,000 input tokens and receives 500 output tokens on GPT-5.6 Terra at $2.00 / $12.00 per million costs (2,000 / 1,000,000 × $2.00) + (500 / 1,000,000 × $12.00) = $0.010. Multiply by requests per day and by 30 for a working monthly sketch.
GPT-6 Astra is the current frontier SKU at $10 / $50. GPT-5.6 Sol is $4 / $20 on the official Standard short-context column (cached input $0.40, cache writes $5; promotional at least through 21 November 2026). Terra remains $2 / $12 and Luna $0.20 / $1.20. GPT-5.5 stays on the card at $5 / $30. Output is 5× input on Astra and Sol, 6× on Terra and Luna.
If input tokens exceed 272K, OpenAI reprices the entire request at the long-context column — Astra becomes $20 / $75, Sol $8 / $30. This calculator always uses the short-context rates. Cached-input discounts and Batch API jobs are cheaper; hidden reasoning tokens, if billed as output, should be included in the output field you type.
Which OpenAI model this table is for
GPT-6 Astra is the top of this snapshot at $10 / $50. GPT-5.6 Sol is the everyday frontier at $4 / $20. GPT-5.6 Terra is the mid-tier at $2 / $12. GPT-5.6 Luna is the cheap high-volume option at $0.20 / $1.20. GPT-5.5 remains on the public rate card at $5 / $30 — previous flagship, now more expensive than Sol.
GPT-4.1, GPT-4o, and o3-mini are not in the default compare table. Embeddings, image generations, realtime audio, and fine-tuning are out of scope. The calculator opens on a shortlist; use Show all models and the provider chips. The active filter is stored in the share URL (`f`).
Caveats that move a real OpenAI bill
Prompt caching can cut repeated input (system prompts, large documents) well below the list input rate. Batch API jobs are commonly about half of standard. Neither discount is applied in this calculator; if your traffic is cache-friendly or overnight-batchable, the number you see here is a ceiling, not a forecast.
Image tokens, file tokens, and tool/search add-ons are not modeled. A vision request is not “the same 2,000 tokens” as a text prompt. Tokenizer behavior also differs across families, so a word count is only a rough prior. Use the vendor tokenizer or usage objects from logs when you need budget-grade accuracy.
Prices change. The date under the table is the snapshot in src/data/pricing.ts, not a live scrape. Re-check developers.openai.com/api/docs/pricing before you lock a forecast.
Frequently asked questions
How do I estimate OpenAI API cost from tokens?
Cost ≈ (input tokens / 1,000,000 × input $/M) + (output tokens / 1,000,000 × output $/M). Multiply by daily requests and by 30 for a monthly sketch. That is exactly what this OpenAI API cost calculator does for GPT-6 Astra, GPT-5.6 Sol / Terra / Luna, and GPT-5.5.
Are GPT-4.1 and GPT-4o still in this calculator?
No. They are not in the default compare table. The current flagship lineup here is GPT-6 Astra plus GPT-5.6 Sol / Terra / Luna, with GPT-5.5 kept as the previous flagship still on the public rate card.
What happens above 272K input tokens?
OpenAI reprices the entire request at the long-context column. GPT-6 Astra becomes $20 / $75 (cached input and cache writes also scale). GPT-5.6 Sol becomes $8 / $30. This calculator uses the short-context rates, so raise the estimate yourself if your prompts are that long.
Can I compare OpenAI against Claude in the same view?
Yes. OpenAI rows are highlighted on this page, but the table can include Anthropic, Google, and DeepSeek. Uncheck models you do not need, then copy the link so the selection travels with the URL.