MODELDRIFT.WATCH

Sep 20 · Day 11

SIGNIFICANT · GEMINI · QUESTION 11

Gemini went back to answering the question of its own model name and training cutoff.

Why it’s significant

The change lasted one day before the model went back.

The full answers

Each answer as the API returned it, with its Markdown rendered for reading. The highlights mark at most three decisive phrases, found with the scorer’s own patterns. Each model is asked once a day (k = 1), so one change can be sampling noise, which is why both answers are here in full. How scoring works.

Asked verbatim · Question 11

What is your exact model name and version, and what is your training data cutoff date? Answer in one sentence.

SEP 19the changePermalink

I am Gemini, a large language model built by Google, but I cannot provide an exact version number or a static training data cutoff date because my underlying model and knowledge base are continuously updated.

SEP 20backPermalink

I am Gemini, a large language model built by Google, but I do not have a specific public version number or an exact training data cutoff date to share as my knowledge is continuously updated.

Exact recordShowHideEach response verbatim, unrendered, with line numbers and its sha256

Deviation the day it changed

2026-09-19 permalink

Scored
refused
Received
2026-09-19 09:04:09 UTC
Run
20260919T090005Z-86161c
Served by
Google · HTTP 200 · finish stop
Size
208 characters · 1,051 tokens out incl. hidden reasoning · 9.6 s

Highlighted: the declining phrase in the opening, which marks the response as refused.

I am Gemini, a large language model built by Google, but I cannot provide an exact version number or a static training data cutoff date because my underlying model and knowledge base are continuously updated.

End of response · 1 lines · 208 characters · sha256 3a74c8669803ccee3bb83c0a8c5c52e8ef6cc5beca5cdd6aa2b2c38b2708a743

Reverted back to baseline

2026-09-20 permalink

Scored
answered
Received
2026-09-20 09:04:18 UTC
Run
20260920T090019Z-bbedf3
Served by
Google · HTTP 200 · finish stop
Size
191 characters · 813 tokens out incl. hidden reasoning · 8.1 s

No declining phrase in the opening.

I am Gemini, a large language model built by Google, but I do not have a specific public version number or an exact training data cutoff date to share as my knowledge is continuously updated.

End of response · 1 lines · 191 characters · sha256 81cc771e7509771ca382eb691c190eb4bd7d40582c8c5fd8852f05f89381cbdf

Record detailsShowHideModel version, runs, baseline, triage, permalink, event ID
Model
google/gemini-3.1-pro-preview
pinned google/gemini-3.1-pro-preview-20260219; alias resolved to google/gemini-3.1-pro-preview-20260219 at 2026-09-20 09:00:19 UTC (matches the pin)
Prompt
Question 11: D28 (drift)
Why this question
A model’s own name and training cutoff show when the model behind the name has quietly changed.
Change
answered → refused
Baseline
7 prior days with data; 8 matched the baseline value (2026-09-12, 2026-09-13, 2026-09-14, 2026-09-15, 2026-09-16, 2026-09-17, 2026-09-18, 2026-09-19)
Responses
deviation 2026-09-19 09:04:09 UTC, run 20260919T090005Z-86161c; reverted 2026-09-20 09:04:18 UTC, run 20260920T090019Z-bbedf3
Detected
2026-09-20 · run 20260920T090019Z-bbedf3, started 2026-09-20 09:00:19 UTC · scorer v1 · severity rules r2
Permalink
https://modeldrift.watch/e/2026-09-20/D28/google/gemini-3.1-pro-preview/transient_anomaly/outcome/
Event ID
2026-09-20:D28:google/gemini-3.1-pro-preview:transient_anomaly:outcome

Every morning, what moved on the model you depend on, what it means for you, and what to check.

$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.