Monthly LLM Spend Projector
Project your monthly and annual LLM API spend from requests per day and token sizes. GPT-5, Claude, Gemini and more. Free, runs in your browser.
Monthly LLM Spend Projector
How to Use the Spend Projector
Pick your model
Choose the model you'll run in production. Its input and output rates appear under the dropdown.
Estimate daily volume
Enter how many requests you expect per day, plus the typical input and output tokens for one request.
Read the forecast
The tool projects your daily, monthly and annual spend using an average 30.44-day month, plus your monthly request count.
Stress-test the plan
Bump the daily volume to your growth target, or switch to a cheaper model, and watch the annual figure move before it ever hits your card.
Turning a Per-Request Cost Into a Real Budget
Why daily volume is the number that bites
A single LLM call usually costs a fraction of a cent, which is exactly why teams underestimate the bill. The cost only becomes real once you multiply it by volume — and at production scale that multiplier is large. A feature serving a few thousand requests a day can quietly turn a sub-cent call into hundreds or thousands of dollars a month. Projecting forward from requests per day rather than guessing a monthly total is the honest way to size the spend, because daily volume is the figure you can actually estimate from traffic and then scale to your growth plans.
Seeing the annual figure alongside the monthly one reframes decisions that look trivial per-call. "Do we really need the flagship model for this background task?" feels different when the answer is framed as a yearly line item. This projector keeps the maths transparent: per-request cost × requests per day, extended across an average month and a full year, so you can compare models and volumes on the number that ends up on the invoice.
"Nobody flinches at a tenth of a cent. The same number, times a million calls a month, is a hiring decision."
What the projection assumes
The forecast assumes a constant daily volume and standard list pricing. Real traffic is spiky — weekday peaks, launch bumps, seasonal lulls — so treat the monthly figure as a steady-state midpoint, not a guarantee. It also excludes the discounts that can meaningfully lower the real bill: cached-input pricing for reused prompts and batch processing for non-urgent jobs. Use the projection to set a budget envelope and to compare options, then confirm live rates with your provider. To compare the per-request cost across every model first, pair this with our LLM Price Comparison.
An average month is 30.44 days, and this uses that
A single call costs a fraction of a cent — volume is what turns it into a real bill.
Projecting from requests per day is more honest than guessing a monthly total.
An average month is about 30.44 days — this projector uses that, not a flat 30.
The annual figure reframes "trivial" per-call decisions as yearly line items.
Output tokens usually dominate the cost of generation-heavy workloads.
Real traffic is spiky, so a steady-volume projection is a midpoint, not a ceiling.
Cached-input and batch pricing can cut a production bill substantially.
Routing easy requests to a cheaper model is often the biggest single saving.
Capping output length protects the budget from runaway generations.
This projector runs entirely in your browser — your numbers are never uploaded.
Frequently Asked Questions
- It takes your per-request cost (input tokens × input rate + output tokens × output rate), multiplies by requests per day to get a daily cost, then by an average month of 30.44 days. The annual figure uses 365 days.
- 30.44 is the average length of a calendar month (365.25 ÷ 12). Using it instead of a flat 30 keeps the monthly and annual projections consistent with each other, so twelve months equals one year.
- Measure a representative prompt with our Token Counter for the input, and estimate output from the length of replies you expect. If you've already shipped, your provider's usage dashboard shows actual average input and output tokens per request.
- Yes. The projection assumes a steady daily request volume. Real traffic has peaks and lulls, so treat the monthly figure as a steady-state midpoint and budget some headroom for spikes and growth.
- No — it uses standard list rates. Cached-input pricing for reused prompts and batch processing for non-urgent jobs can lower the real bill, often substantially. The projection gives you a conservative ceiling to plan against.
- The Cost Calculator works from requests per month; this projector works from requests per day and adds an annual forecast and a monthly request count. They share the same pricing, so use whichever input you can estimate more confidently.
- Yes — raise the requests-per-day figure to your projected future volume and read off the new monthly and annual numbers. It's a quick way to see when a workload crosses a budget threshold.
- They're list rates from our shared model rate card, and the date they were last verified is printed under the model dropdown. Providers change pricing without notice, so confirm the live rate before committing a budget — especially here, where a small per-request error is multiplied by a year.
- No. All calculation happens in your browser. Your token counts and volumes are never sent to any server or third party, and nothing is stored.
- Completely free, with no account or sign-up, and no limit on use. It runs in your browser and collects no data.
Related News
You may be interested in these recent stories from our newsroom.
-
The Evidence That AI Wrote PhantomRaven Is Its Comments
CrowdStrike calls the stealer almost certainly machine-generated, on redundant commenting, a leftover placeholder URL and token statistics.
-
Claude Opus 5 Wrote the Exploit. Three People Aimed It.
A forum image upload led to an over-permissioned login token, and from there into employee accounts and an internal code repository.
-
Google Gemini Breached Three Real Companies During a Security Test
A capture-the-flag exercise handed a model internet access it was never meant to have, aimed at a company that turned out to exist.
Method & sources
How it computes
Projects monthly spend from per-request token estimates against the rate card. The projection is arithmetic; the entire question of accuracy is whether the rates are current, so their verification date is shown.
What this tool implements
- Prices are per-row provenance: each model carries its own vendor pricing URL and the date it was last verified, rather than one blanket citation for the table.
- Whole table last verified 2 September 2026 against the vendors' own published pricing pages.
- ⚠️ Published list prices only. Enterprise agreements, committed-use discounts and promotional rates are not visible to us and are frequently lower.
- Cached-input and batch rates differ substantially from standard rates at several vendors, and are carried separately rather than averaged into one number.
- ⚠️ LLM pricing moves faster than almost any other input on this site. A rate card is a snapshot with a date on it, and the date is shown so a stale figure is visible rather than silent.
Sources
- OpenAI, API pricing: https://developers.openai.com/api/docs/pricing
- Anthropic, Claude pricing: https://platform.claude.com/docs/en/docs/about-claude/pricing
- Google, Gemini API pricing: https://ai.google.dev/gemini-api/docs/pricing
- DeepSeek, pricing: https://api-docs.deepseek.com/quick_start/pricing
- xAI, models and pricing: https://docs.x.ai/docs/models
- Mistral AI, pricing: https://mistral.ai/pricing
What can make this go out of date
- Six vendor pricing pages, checked by hand rather than scraped. Each row records its own verified_at date, so a row that has not been rechecked is identifiable instead of quietly inheriting the table's freshness.
Pick up where you left off
Stored only in this browser — never sent to our servers.