tool / ai
Enter your expected token counts and compare real-time API costs across 300+ models from OpenRouter — GPT-4o, Claude, Gemini, Llama, Mistral, and hundreds more. Prices fetched live, no API key required.
← Back to tools| Model | Context | Input/1M | Output/1M | Input cost | Output cost | Total |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Model prices are fetched directly from the OpenRouter public API at page load. OpenRouter aggregates pricing from OpenAI, Anthropic, Google, Meta, Mistral, and many other providers — so you get one place to compare them all.
Enter your expected input tokens (your prompt + context) and output tokens (the model response). Use the search box to filter by model name or provider. Click any column header to sort. The "Total" column is your estimated cost per API call.
Not sure how many tokens your text is? Use the Token Counter tool to count tokens in your prompt before estimating costs. As a rough guide: 1,000 English words ≈ 750 tokens.
Model pricing spans roughly three orders of magnitude. The same 2,000-token prompt that costs a fraction of a cent on a small open-weights model can cost 100× more on a frontier model — and for plenty of jobs (classification, extraction, routing, summarizing short text) the cheap one is good enough. Most teams pick a model once, hardcode it, and never look at the bill until it hurts. Five minutes here before you commit is the cheapest optimization you'll ever do.
Budgeting a batch job. You've got 50,000 documents to summarize. Multiply your per-call total by 50,000 and you know whether the job costs lunch money or a month's cloud budget — before you run it.
Sanity-checking a model switch. Someone on the team wants to upgrade to the newest flagship model. Type in your real traffic numbers and see what the upgrade actually costs per call, not what the marketing page implies.
Finding the floor. Sort the Total column ascending and work upward. Start evaluating quality from the cheapest model that plausibly handles your task, not from the most famous one.
No. The pricing endpoint is public. The page fetches the model list anonymously when it loads — nothing about you is sent anywhere.
They're whatever OpenRouter reports at the moment you load the page. Refresh to re-fetch. Prices are OpenRouter's rates, which usually track the provider's direct pricing closely but aren't guaranteed identical.
Two reasons: your real token counts probably differ from your estimate, and features like prompt caching, batch discounts, or tiered pricing aren't modeled here. Treat the numbers as a comparison tool, not an invoice predictor.
Typically 2–5× the input rate, and it varies by model. That's why the calculator asks for both numbers separately: a chatty model with long responses can cost more than a pricier model that answers tersely.
// huntermussel
HunterMussel designs cost-efficient AI pipelines with intelligent model routing, prompt caching, context compression, and fallback strategies — so you scale without the runaway API bills.
Explore AI Automation Services →Advertisement · Publicidade