About TokenForge

TokenForge is a free LLM Token Cost Calculator built around one idea: the simplest jobs should not need an account, an upload, or a wait.

Operated by bsgoods · the Netherlands · Last updated July 2026

What it does

TokenForge compares the two ways of paying for large language models. On one side, pay-per-token APIs: it tracks input and output prices per million tokens for OpenAI, Anthropic, Google, DeepSeek and hosted open-weights models. On the other side, self-hosting: it converts the rental price of a GPU server, its throughput in tokens per second and a utilization estimate into an effective cost per million tokens. You enter your monthly volume and input:output ratio; it returns both monthly bills, the break-even volume, and a capacity check that tells you whether a single server can even produce the tokens you need.

Why we built it

Provider pricing pages are scattered, aggregator tables go stale within weeks, and most API-versus-self-hosting comparisons quietly skip the numbers that decide the outcome: throughput, utilization, and the operations time a private deployment really costs. We wanted one page with dated, citable prices and the whole comparison done honestly — including the cases where self-hosting never wins.

How it works

Everything runs in your browser — nothing you type is uploaded or stored. The page loads a single prices.json file from our own domain. That file is regenerated daily by an automated pipeline that reads public sources (the community-maintained LiteLLM price table and the OpenRouter models API), rejects suspicious price jumps for manual review, and stamps the data with the date you see in the widget.

How we calculate

The method is deliberately simple and fully visible:

• blended $/1M = (ratio × input price + output price) ÷ (ratio + 1) × tokenizer factor
• API cost/month = tokens per month × blended $/1M
• self-host cost/month = server price/month × ops overhead
• server capacity = tokens/sec × utilization × 2,592,000 seconds
• break-even = self-host cost ÷ blended $/1M

The tokenizer factor corrects for models that split the same text into more tokens (Anthropic's newest models produce roughly 30% more), so cross-provider comparisons stay fair. Prices shown exclude batch and prompt-caching discounts — those can cut an API bill by half or more, and the FAQ on the calculator page explains when they apply.

Who it's for

• Developers — estimate the API bill before shipping an AI feature.
• Startups and product teams — find the volume where self-hosting starts to pay off, and budget with defensible numbers.
• Self-hosters — check whether a rented GPU really beats the API at your usage.
• Anyone who just needs a quick answer.

Who runs TokenForge

TokenForge is built and maintained by the TokenForge team, an independent project operated by bsgoods, based in the Netherlands. It is not affiliated with the platforms or products it references; those names and specifications belong to their respective owners and are provided for guidance only. Spotted a mistake or have a request? Get in touch — the same people who build the tool read the messages.

Privacy and ads

TokenForge has no accounts and collects no personal data. It is free to use, and the running costs are covered by advertising through Google AdSense. We use Google Consent Mode, so advertising cookies stay disabled until you choose to accept them. The full details are in our privacy policy.

Learn more

New to the topic? Start with our guides. Got feedback or a request? Get in touch.