OpenAI API Cost Calculator
How much will the OpenAI API cost you per month? This page is the TokenForge calculator preset for exactly that question: pick a GPT tier, enter your monthly token volume and input:output ratio, and read off the bill. The preset starts on GPT-5.4 Mini — the workhorse tier most production traffic actually runs on — and prices refresh daily.
OpenAI API cost tool
At 100M tokens/month, the API is cheaper by $1,959.70/month.
† Anthropic's newest models (Fable 5, Opus 4.8, Sonnet 5) tokenize the same text into ~30% more tokens; the calculator corrects for this. Prices exclude batch (−50%) and prompt-caching discounts.
What OpenAI costs per month (July 2026)
| OpenAI model | Input $/1M | Output $/1M | Blended @4:1 | 10M/mo | 100M/mo | 1B/mo |
|---|---|---|---|---|---|---|
| GPT-5.6 Sol | $5.00 | $30.00 | $10.00 | $100 | $1,000 | $10,000 |
| GPT-5.6 Terra | $2.50 | $15.00 | $5.00 | $50 | $500 | $5,000 |
| GPT-5.6 Luna | $1.00 | $6.00 | $2.00 | $20 | $200 | $2,000 |
| GPT-5.4 Mini | $0.75 | $4.50 | $1.50 | $15 | $150 | $1,500 |
| GPT-5.4 Nano | $0.20 | $1.25 | $0.41 | $4.10 | $41 | $410 |
Monthly figures assume a 4:1 input:output ratio — four prompt tokens for every token the model writes back, which is typical for chat and assistant workloads. The spread matters more than any single price: the same traffic costs 24× more on GPT-5.6 Sol than on GPT-5.4 Nano.
How the bill is calculated
blended $/1M = (ratio × input price + output price) ÷ (ratio + 1)monthly bill = tokens per month (M) × blended $/1M
Worked example. A support chatbot handling 20,000 conversations a day at roughly 800 input and 200 output tokens each uses about 600M tokens/month at a 4:1 ratio. On GPT-5.4 Mini that is 600 × $1.50 = $900/month. The same traffic on GPT-5.6 Terra costs $3,000, and on GPT-5.6 Sol $6,000 — tier choice is by far the biggest lever on an OpenAI bill.
Picking the right GPT tier — and cutting the bill
Match the tier to the task, not to the demo. GPT-5.6 Sol is the flagship for the hardest reasoning; Terra is the mainstream default for user-facing features; Luna and GPT-5.4 Mini carry high-volume workloads like summarization and drafting; and GPT-5.4 Nano at $0.41 blended is built for classification, extraction and routing. Most products end up mixing tiers — an expensive model for the few calls that need it, a cheap one for the many that don't.
Take the discounts the price list doesn't show. The Batch API cuts prices by about half for anything that can run asynchronously — nightly summaries, embeddings backfills, evaluation runs. Prompt caching makes repeated input tokens roughly 90% cheaper (on Terra, $0.25 instead of $2.50 per 1M), which is substantial for chatbots that resend a long system prompt with every message. The calculator shows undiscounted list prices, so its result is your ceiling, not your floor.
Get the volume number right before optimizing the rate. A price per million tokens means nothing until you know how many millions you use. If you're still guessing, work through how to estimate your monthly LLM token usage first — requests per day and words per prompt pin the number down in a few minutes. Note that input-heavy workloads like RAG can run at 10:1 or more, which pulls the blended price down toward the cheaper input rate.
At very high volume, check the exit. From about 1.41B tokens/month, a rented H100 running an open-weights model undercuts even GPT-5.4 Mini at the preset assumptions — and it beats Terra from about 422M. If your bill is heading that way, the OpenAI vs self-hosting comparison has the full break-even table per tier.
Frequently asked questions
Related tools and guides
- LLM Token Cost Calculator — the main tool, with all providers and servers
- OpenAI vs self-hosting cost — the break-even table per GPT tier against a rented H100
- Claude API Cost Calculator — Anthropic pricing, including the ~30% tokenizer correction
- Gemini API Cost Calculator — Google's tiers from budget Flash-Lite to Gemini 3.1 Pro
- How to estimate your monthly LLM token usage — get the volume number this bill depends on