That’s the free tier request-per-minute rate. Since your tier showing at the top means you didn’t accidentally pick the wrong org, looks like something broke for the organization or the backend. Same here.
Are they enforced? No, but the API latency is hella slow even just to the first token out.
Where it should be at for gpt-4o by model page:
– Fingerprint report from 100 trials:
gpt-4o: (28):fp_6dd05565ef, (2):fp_726d488742, (70):fp_898ac29719
For 10 trials of gpt-4o @ 2025-04-02:
| Stat | Average | Cold | Minimum | Maximum |
|---|---|---|---|---|
| stream rate | Avg: 60.350 | Cold: 68.6 | Min: 38.7 | Max: 82.4 |
| latency (s) | Avg: 3.053 | Cold: 0.9364 | Min: 0.4859 | Max: 22.5931 |
| total response (s) | Avg: 4.165 | Cold: 1.8548 | Min: 1.2884 | Max: 24.2137 |
| total rate | Avg: 32.536 | Cold: 34.505 | Min: 2.643 | Max: 49.674 |
| response tokens | Avg: 64.000 | Cold: 64 | Min: 64 | Max: 64 |
Mitigations have been put in place before when things all went south in the past, but nothing reported about the API.
