Searches like which AI API has no rate limits usually mean "won't throttle my product during a demo." Every provider limits something — the question is whether limits are transparent tokens/RPM caps or fair-use on a flat plan.
Types of limits
| Limit | Meaning |
|---|---|
| RPM | Requests per minute |
| TPM | Tokens per minute |
| Concurrency | Parallel streams |
| Daily quota | Hard cap on free tiers |
| Fair use | Soft cap on "unlimited" plans |
Pay-per-token throttling
Token vendors stop you with 429 Too Many Requests when you exceed tier quotas. You can raise limits by spending more — but spikes still hurt if finance did not expect it.
Flat-rate fair use
Flat-rate APIs advertise predictable price; capacity is shared. Fair-use policies prevent abuse while letting normal products breathe. That is different from per-token panic but not the same as infinite.
Daymora publishes plan terms on pricing.
Designing around limits
- Exponential backoff on 429 (retries guide)
- Queue heavy jobs offline
- Shard keys only when terms allow — not a loophole
"Unlimited" marketing
Consumer chat "unlimited" is fair-use gated. API "unlimited tokens" on flat plans still means reasonable production traffic, not cryptocurrency mining.
Compare billing shapes: Flat-rate vs pay-per-token.
Free tiers are the tightest
See Best free tier comparison.
Bottom line
No reputable AI API is limit-free. Choose limits you can design for: token RPM/TPM meters for spiky experiments, fair-use flat plans for steady products — and read the doc page before you promise customers infinite AI.