Fair use is the whole point of a free tier. Everyone gets the same daily allowance, and nobody should be able to quietly take far more than their share. This week we found and fixed a bug that broke that promise, corrected the accounts it affected, and tightened the free lane so it stays fair for everyone. This post explains exactly what was wrong, who was affected, what we changed, and what to expect now.
The bug, in plain terms
Our free models are billed at $0 per token — that is the whole design. To make that work, the gateway set those models’ price ratio to zero, which is what makes the free lane free. The side effect was the bug: because the price was zero, every request on a free model billed $0 against your balance, no matter how many tokens it actually used. The prepaid balance never moved, no matter how much you consumed.
Requests were still limited, but only by a per-account cap on the number of requests — and not on the size of each one. So a single account could send request after request, each returning a very large response, and its balance would read the same at the end as at the start. The daily request counter went up. The token volume behind those requests was effectively unlimited. One heavy account could consume millions of tokens in a day while its balance sat at zero the entire time.
To be clear about who this affected: if you used the service normally, nothing was wrong with your account and you did not owe anything. The issue was volume. A small number of accounts pushed the free lane extremely hard — automated scripts and very large prompts — and those are the accounts the fix touched. Normal chat and API use was never in danger of this.
What we changed
We fixed the calculation so usage is counted correctly and capped, on a per-account daily basis. As of 2026-11-13, around 02:07 UTC (07:37 IST), every account has a clear, equal daily free allowance:
- 1,000 requests per day on free models (raised from the old 100), and
- 5,000,000 tokens per day on free models.
Either cap pauses free-model use until the next reset. Both counters now track real token volume, so a very large response counts against your allowance the same as a small one. The counters reset together at 00:00 UTC (05:30 IST) every day, automatically — no action needed. Your balance and the free lane daily allowance are the two things that decide whether free requests go through, and both are shown live on your console overview with progress bars for requests and tokens.
This was not a price increase and not a plan change. The update was only about fixing how balance and usage were calculated. Free models are still $0 per token, still need no card, and still require only a free API key. Every user is free to use their full daily quota, exactly as before — the allowance just can no longer be silently exceeded.
The cleanup and the reset
Fixing the calculation forward was only half the job. Accounts that had already consumed large volumes under the buggy accounting needed to be brought back in line, so we ran a one-time correction at the same time as the fix:
- 438 accounts were corrected. Their balances were adjusted to reflect the request and token volume they had actually consumed under the old accounting.
- In total about 17,590 requests were re-accounted, correcting roughly $1,059 of usage that had not been counted.
- 112 accounts had their balance brought to $0 rather than into a negative figure — we chose not to leave anyone owing money over this. Nobody’s balance was pushed below zero by the correction, and nobody lost access to their daily free allowance.
The correction ran at the same moment the caps went live, so from that point on every account is on the same fair footing. If your balance changed on 2026-11-13 and you were a heavy free-lane user, that is this correction — and it does not affect your daily free allowance, which is intact for every account.
We also removed the models that kept hitting limits
While we were in there, we looked at the free model list itself. A handful of upstream models were hitting their shared daily cap so often that calling them was mostly a waste of your time — you would pick one, get a limit error, and have to try again. We have removed those frequently-failing models from the fetch-models and available-model lists so you stop calling them at all.
That means the model catalog and the free models list now only show free models that reliably answer. If a model is listed, it is expected to work; if one is having a bad day, the status board shows it in real time and the gateway falls back to a model that always answers rather than returning you a dead end. Removing the flaky entries also stops clients from repeatedly retrying a model that cannot serve them, which was adding to the load on everyone else’s allowance.
What this means for you
- Your daily free quota is intact and fair. 1,000 requests and 5M tokens a day, free, no card, resetting at 05:30 IST.
- Usage is counted honestly now. Large responses count against your allowance, so nobody can take more than their share quietly.
- The model list is cleaner. Models that frequently hit limits are gone from the lists, so you stop calling models that cannot answer.
- Paid models are unaffected. If you add credit and use a paid model, everything works as documented in pricing and the rate limits.
We would rather fix a fairness bug and be honest about it than leave it in place. If you have questions about your account specifically, check your console usage first — it now shows your live request and token allowance — and watch the changelog for follow-up improvements to the free lane.
Watch the changelog for the next additions.