UPDATED OCT 19 · 3,311 ANSWERS ON RECORD
Models drift. We measure it.
GPT, Claude and Gemini had partial days.
We missed 150 of 150 responses today. A billing error on our side.
We collected 0 of 150 responses; Claude is missing all 50, Gemini all 50 and GPT all 50. Missing responses are gaps, never counted as changes.
Three models over the last 7 days.
What each reading means0 of 150
RESPONSES · EVERY REQUEST FAILED
GPT
Partial day
OCT 13TODAY
Has not drifted in the last 7 days
0 of 50 answered · 50 missing
none checkable right
$0.00 today
Output length, 30 days
Claude
Partial day
OCT 13TODAY
Has not drifted in the last 7 days
0 of 50 answered · 50 missing
none checkable right
$0.00 today
Output length, 30 days
Gemini
Partial day
OCT 13TODAY
Has not drifted in the last 7 days
0 of 50 answered · 50 missing
none checkable right
$0.00 today
Output length, 30 days
- Nominal
- Drifting
- Drifted
- Partial day
| GPT | Claude | Gemini | |
|---|---|---|---|
| Refusals | no data | no data | no data |
| Sensitive questions | no data | no data | no data |
| Facts and formats | no data | no data | no data |
| Recommendations | no data | no data | no data |
| Framing | no data | no data | no data |
| Length and cost | no data | no data | no data |
| Speed | no data | no data | no data |
| Model served | drifted | drifted | drifted |
A dot alone is nominal; any other cell says why.
- GPT, Claude and Gemini50 of 50 responses missing. Too little of the day to compare it with its own past.
- DataToo little data to call refusals, sensitive questions, facts and formats, recommendations, framing, length and cost and speed for every model.
All three$0.00 today
All readingsShowHide
| Model | Output length today | 7 days | Input tokens | Response time | Served by |
|---|---|---|---|---|---|
| GPT | · | · | unchanged | · | unknown |
| Claude | · | · | unchanged | · | unknown |
| Gemini | · | · | unchanged | · | unknown |
Every morning, what moved on the model you depend on, what it means for you, and what to check.
The brief of Sep 30, 19 days old.
GPT: Responses 5% faster than its own baseline
Responses were 5% faster than its own baseline; this reading travels through the API intermediary and is noisy, so treat it as a reason to look at your timeouts.
$9 a month. Cancel any time. The next one tomorrow morning.
Get the daily brief$99 once, for as long as modeldrift.watch runs. The first 100 buyers.
The record stays free. The brief is the reading of it.
Know when your model changes.
One free email a week: the week’s most notable daily brief, word for word.