Free tool
LLM API cost calculator — with the blast radius nobody else shows
Price a workload on every current OpenAI, Claude, and Grok model. Then ask the question the other calculators skip: if this breaks tonight, what does it cost by the time someone notices?
How the math works
A request costs its uncached input tokens at the input rate, plus cached input tokens at the provider's cache-read rate, plus output tokens at the output rate — all per million tokens. Multiply by requests per day for a daily figure and by thirty for a month. The Batch API halves input and output on OpenAI and Anthropic for asynchronous jobs.
The blast radius takes the same workload at a broken request rate — an agent loop, a retry storm, a leaked key — and prices it at each point a team typically finds out: an hourly alert, the next morning, after a weekend, or the monthly invoice. The per-token price barely matters; the detection time is the whole story.
Prices, verified
Every list price was checked on 2026-09-17 against OpenAI, Anthropic, and xAI. It is the same sheet PulseMeteruses to estimate your spend between the provider's nightly cost reports, so when a price changes for our customers it changes here too. The raw data is public at /api/public/prices.json.
Calculators by model
Each page has the model's verified prices, worked examples for common workloads, and its blast radius.
OpenAI
Anthropic
Questions
Where do these prices come from?
From the same price sheet that powers PulseMeter's hourly spend estimates. Every row was checked against the provider's own pricing page on 2026-09-17 (OpenAI, Anthropic, and xAI), and we re-verify it on a schedule because a wrong public price is worse than none.
What is a blast radius?
The money a broken workload burns between the moment it breaks and the moment someone notices. It depends far more on detection time than on the per-token price: the same agent loop costs about fifty dollars if an hourly alert catches it and tens of thousands if the monthly invoice is the first you hear of it.
Does prompt caching apply to output tokens?
No. Cache discounts apply only to input tokens that were served from cache; output tokens are always billed at the full output rate. The calculator's cache-hit percentage therefore only reduces the input side.
Why doesn't the Batch API reduce the runaway numbers?
Because a runaway is never on the batch queue. Batch discounts (50% on OpenAI and Anthropic) apply to deliberate, asynchronous jobs. Loops, retry storms, and leaked keys run through the standard real-time API, so the blast radius tab always prices at standard rates.
Why only OpenAI, Anthropic, and xAI models?
Because those are the providers PulseMeter monitors, so they're the prices we maintain and can vouch for. We would rather show forty models we verify than three hundred we don't.
Are these quotes?
No. They're list-price estimates. Negotiated enterprise discounts, tool-call token overhead, reasoning tokens, long-context tiers (xAI bills double at 200k+ input tokens), and data-residency multipliers all change real bills. Your provider's cost report is the source of truth — which is exactly what PulseMeter reconciles against every night.
Know the hour it breaks, not the month
Connect a read-only key and PulseMeter alerts you in Slack or email the moment spend spikes — on every model you just priced.
Start free, no card