Tokens, credits, minutes and neurons: why free allowances cannot be compared

Every directory that compares free tiers hits the same wall: the units are not interchangeable. Four show up constantly, and each one hides a different trap.

Tokens measure text

A token is roughly three quarters of a word in English. Input and output are both counted, usually at different prices, and a long prompt plus a long answer costs more than the question alone. A million tokens per day is generous for chat and small for summarising documents all day.

Requests have two ceilings

Most APIs publish a daily or monthly request cap and a per-minute rate limit. The lower one decides what you can actually build. Failed and retried attempts often count toward both, so a retry loop can burn a day of allowance in minutes.

Voice credits measure characters

ElevenLabs-style credits convert to characters, not minutes, and the rate changes per model. Other providers bill by the minute of audio, or by the hour of transcription. A plan offering 10,000 credits and a plan offering 185 hours are not comparable without doing the arithmetic for your own files.

Compute credits are money with an expiry

US$200 for 30 days, US$30 per month, or a perpetual always-free allowance are three different products. An expiry is a cliff: unused balance disappears, and the ability to budget disappears with it.

Neurons are weighted by model

Cloudflare's Neurons are a compute unit that varies by model, so 10,000 Neurons might pay for a lot of one model and very little of another. The headline number is meaningless without the model you intend to call.

The only fair comparison

Pick one real task — a page of transcription, fifty image generations, a support bot answering a hundred messages — and measure what it consumes on each provider. Cost per useful output is the number that survives when the marketing page changes.