What will this cost me to run?
Enter your expected traffic and see the monthly bill across every model — before a single user signs up.
Assumes 30 days/month. Cached input billed at each model's own cached rate. See full rates on the pricing page.
How the calculator works
The projection is a straightforward token-volume model. From your inputs we derive monthly request volume and token counts, then price them against every model’s live rates:
- Prompts per month = monthly active users × prompts per user per day × 30 days.
- Input tokens = prompts per month × average input tokens per prompt.
- Output tokens = prompts per month × average output tokens per prompt.
- Monthly cost = (input tokens ÷ 1M) × blended input rate + (output tokens ÷ 1M) × output rate, where the blended input rate applies your prompt-cache hit rate: cached tokens are billed at each model’s own cached-input rate and the rest at the standard input rate.
In other words, a 30% cache hit rate means 30% of your input tokens are billed at the model’s cached rate and 70% at the full input rate. Output tokens are never cached, so they’re always billed at the standard output rate.
A worked example
Say you have 5,000 monthly active users, each sending 8 prompts a day, with 1,500 input tokens and 500 output tokens per prompt, at a 30% cache hit rate. That’s 1.2M prompts/month → 1.8B input tokens and 600M output tokens. For a model priced at $2 input / $10 output / $0.20 cached per 1M, the blended input rate is ($2 × 0.7) + ($0.20 × 0.3) = $1.46/1M, so input ≈ 1,800 × $1.46 = $2,628 and output ≈ 600 × $10 = $6,000 — about $8,628/month. Change any slider and the whole table re-ranks instantly.
If this estimate looks higher than you'd like, four concrete levers move it the most — caching, batching, model routing, and context trimming — each with worked math in How to Cut Your LLM API Bill by 80%. Since caching discounts vary a lot by provider, use the cache savings calculator to see which model rewards your repeated context the most before picking one from this table.
The estimate assumes a 30-day month and steady traffic; real bills vary with burst patterns, retries, and tool-use overhead. Treat it as a planning figure and confirm live rates on the pricing page. See methodology for data sources. This projects chat/completion cost only — if your app also embeds documents for search or RAG, see embedding model pricing and rankings for that side of the bill. And if you're building a multi-step tool-using agent rather than single-turn chat, this model badly underestimates the real cost — use the agent cost calculator instead, since agent context grows quadratically with step count, not linearly.