LEARN / CHATGPT COST PER PROMPT
How much does ChatGPT actually cost per prompt?
Somewhere between “basically free” and “more than you think.” The honest answer depends on how you pay — subscription or API — and on a multiplier almost nobody accounts for. Here is the arithmetic, using mid-2026 list prices.
The subscription math
On a $20/month ChatGPT Plus plan, your cost per prompt is just $20 divided by how many prompts you actually send. A heavy user sending 40 prompts a day (~1,200/month) pays about 1.7¢ per prompt. A casual user sending 5 a day pays about 13¢ per prompt — nearly 8x more per prompt for the same subscription. The subscription is a flat-rate buffet: the economics reward the hungry.
The API math
Pay-per-token pricing makes the cost visible. Take a typical prompt: ~150 tokens of input, ~450 tokens of output. At mid-2026 list prices per million tokens:
| Model | Cost per prompt | 1,000 prompts |
|---|---|---|
| GPT-5.5 ($5/$30 per MTok) | $0.0143 | $14.25 |
| Claude Sonnet 5 ($3/$15) | $0.0072 | $7.20 |
| Claude Haiku 4.5 ($1/$5) | $0.0024 | $2.40 |
| DeepSeek V4 Flash ($0.14/$0.28) | $0.00015 | $0.15 |
Assumes 150 input / 450 output tokens. List prices, no caching or batch discounts.
The multiplier nobody budgets: conversation history
Models are stateless — on every turn, the application re-sends the whole conversation as input. Turn 30 of a chat doesn't cost one prompt; it re-pays for the previous 29. A long working session can make the average prompt cost 5–20x the single-prompt math above. This is the single biggest gap between what people estimate and what shows up on the invoice, and it's why “one long chat for everything” ranks #1 in the Burn Index.
So which is cheaper?
- — Heavy interactive use (dozens of prompts daily): the subscription almost always wins. Flat rate, no meter anxiety.
- — Light use: the API is dramatically cheaper — a casual user's ~150 prompts/month costs ~$2 on GPT-5.5, a tenth of the subscription.
- — Anything automated (scripts, agents, pipelines): API only — and now volume, model choice, and prompt hygiene are your bill. Measure before you scale.
Want the number for your actual prompt instead of an average? Paste it into the Burnmeter — tokens, cost across nine models, and a verdict on how much of it is ceremony.