allocateLLM price comparison
Enter your monthly requests, token sizes, and cache hit rate. See what the same workload costs on every model, cheapest first.
Same workload, 199x gap between Llama 4 70B and Opus 4.8. Route each task to the model that fits it.
How it works
Common questions
Where do the prices come from?
Published per-million-token API prices for each model, split by input and output tokens, with cached input priced at roughly 10% of the input rate. Prices move, so check the exact rate before you commit a budget.
Why do output tokens cost more?
Generation is the expensive direction: every output token requires a full forward pass, while input tokens are processed in parallel. Most providers price output 4 to 5 times above input.
What is a prompt cache hit?
When the start of your prompt (system instructions, tool definitions) repeats across requests, providers can reuse the computed state and charge around a tenth of the normal input price for those tokens. Agents with long stable system prompts see high hit rates.
Do I have to pick one model?
No. Most production teams route: judgment-heavy traffic goes to a frontier model, high-volume routine traffic to an open-weight one. On Allocate each route names its model, so you can swap either without a deploy when prices or quality move.
Is this calculator free?
Yes, free and unlimited, no account. It runs in your browser.
More free tools
These are list prices. On Allocate every model here sits behind one key with per-token metering, hard caps, and a bill you forecast before you spend.