OpenAI API Pricing Explained: What GPT-5 Actually Costs
OpenAI prices the GPT-5 family per million tokens, billed separately for input (what you send) and output (what the model writes). This guide breaks down current list prices, real worked examples, and the discounts that actually move the needle.
GPT-5 API Prices (Per Million Tokens)
As of October 2026, standard list pricing for the main GPT-5 tiers:
- GPT-5.5 (flagship) — $5.00 input / $30.00 output per 1M tokens
- GPT-5.4 (mid-tier) — $2.50 input / $15.00 output per 1M tokens
- GPT-5.4-Nano (budget) — $0.20 input / $1.25 output per 1M tokens
Output is 4–6× the input price across the board, because generating text token-by-token costs far more compute than reading it. OpenAI updates prices over time — confirm on OpenAI's pricing page before you commit to a budget.
Worked Examples: What Real Usage Costs
One 1,000-word article (≈1,333 input tokens + 1,333 output tokens) on GPT-5.4: (1,333 ÷ 1M × $2.50) + (1,333 ÷ 1M × $15.00) ≈ $0.023.
A chat app serving 10,000 messages/day, averaging 2,000 input + 500 output tokens each on GPT-5.4:
- Daily input: 20M tokens × $2.50/1M = $50.00
- Daily output: 5M tokens × $15.00/1M = $75.00
- Monthly ≈ $3,750 — before discounts
Swap the same workload to GPT-5.4-Nano and it drops to roughly $375/month — a 10× difference for tasks a budget model handles fine. Model your own numbers in the AI Token Counter, which supports GPT pricing out of the box.
Discounts That Actually Matter
- Prompt caching — repeated input (system prompts, document prefixes) can cost about 90% less when cached. For agents that resend a large system prompt every turn, this is the single biggest lever.
- Batch API — 50% off for asynchronous jobs you can run within 24 hours: overnight pipelines, bulk classification, evaluations.
- Model routing — send simple tasks (classification, extraction, rewriting) to Nano and reserve GPT-5.5 for the few calls that need frontier reasoning.
How to Estimate Costs Before You Build
Three steps keep surprises out of your invoice:
- Count tokens in your real prompts and expected outputs — paste them into the token counter instead of guessing from word counts.
- Multiply by planned call volume, using the per-tier rates above. Always budget on output tokens first — they dominate the bill.
- Compare against alternatives: Claude and Gemini mid-tiers are priced similarly, and DeepSeek undercuts all of them for high-volume work. See the full GPT vs Claude vs Gemini comparison.
Estimate your GPT, Claude, and Gemini costs instantly.
Try AI Token Counter →Frequently Asked Questions
How much does 1 million tokens cost on the OpenAI API?
It depends on the model: GPT-5.5 costs $5 input and $30 output per million tokens, GPT-5.4 costs $2.50 and $15, and GPT-5.4-Nano costs $0.20 and $1.25. Input and output are billed separately.
Why are output tokens more expensive than input tokens?
Input tokens are processed in one parallel pass, while output tokens are generated sequentially — one full model pass per token. That makes generation roughly four to six times more expensive than reading.
How much does a ChatGPT-style app cost per month?
For 10,000 messages a day averaging 2,000 input and 500 output tokens, expect roughly $3,750/month on GPT-5.4 at list prices — or about $375 on GPT-5.4-Nano, with caching and batch discounts cutting it further.
How can I reduce my OpenAI API bill?
Use prompt caching for repeated context (up to ~90% off input), run non-urgent work through the Batch API (50% off), route simple tasks to cheaper tiers like GPT-5.4-Nano, and cap responses with max_tokens.