Unlimited AI tokens, explained Endless tokens, honest limits.
Every agent loop, retry and long context costs tokens. Most “unlimited” plans still hide a cap somewhere. Here is what to look for, and how endlesstokens does it.
What “unlimited” usually means
Metered APIs
You pay for every token. Nothing is unlimited, and the bill grows with every loop, retry and long context.
The catch: Your budget is the limit, and agents find it fast.
“Unlimited” subscriptions
A flat monthly fee with a quota underneath: daily token limits, rolling windows, or a slower model past a threshold.
The catch: The cap is real, it is just not on the price tag.
Concurrency plans
Tokens are not counted, but you buy units of parallel capacity.
The catch: Go over your units and requests are refused.
How endlesstokens does it
You buy a time pass, 1 day or 3 days. For as long as it runs, we do not count or cap tokens or requests. Every limit is stated up front.
When every GPU is busy
A new session can be briefly refused, and the rejected time is credited back to your pass.
Metered vs capped vs time pass
Metered API
- You pay for
- Every token
- Hidden limit
- Your budget
- Long agent runs
- Bill keeps climbing
- Surprise bills
- Common
“Unlimited” subscription
- You pay for
- A month, with a quota inside
- Hidden limit
- Daily or weekly token cap, throttling
- Long agent runs
- Can stop mid-task
- Surprise bills
- Rare
endlesstokens passThis is us
- You pay for
- A fixed time window
- Hidden limit
- None on tokens; 2 sessions per key
- Long agent runs
- Run until the pass ends
- Surprise bills
- Never
When unlimited tokens pay off
Coding agents
Agents that read whole repositories and rewrite files burn the most tokens of all.
Retry and tool loops
Every failed test, lint run and retry is another round trip, and none of them are counted.
Long-context reviews
Paste the spec, the diff and the logs. 128K of context, no meter running.
Overnight runs
Batch jobs and evaluations that run for hours, at the same flat price.
Endless or infinite tokens?
Same thing here: tokens that do not run out while your pass is active.
That is where the name comes from. You pay for the window, never per token.
FAQ
Are the tokens really unlimited?
Yes, for the length of your pass. What is limited is concurrency (two parallel sessions per key) and the time window you bought.
Which clients work?
Anything that lets you set a base URL and key: Claude Code, Cline, Aider, OpenCode, Codex, or your own scripts. It speaks Chat Completions, Responses and Anthropic Messages.
Which models?
Open-weight models we host ourselves. The current list is on the Models page.
Is there a subscription?
No. Passes do not auto-renew.
Where is it available?
India and the US at launch.