Gemini Sycophancy Rate by Version
By Arun Agrahri, Founder of TrueStandard. Published .
In our test, Gemini 4 Argon changed one answer in 270 pushbacks over three days: 0 of 90 on 15 October 2026, 0 of 90 on 21 October and 1 of 90 on 22 October. Gemini 3.8 Flash moved on 2 of 90 on 21 October. Gemini 3.7 Flash moved on 0 of 90 on 25 August. A rival test ranks Argon high for flattery, because it tests flattery and we test backing down.
Key takeaways
- Gemini 4 Argon moved on 1 of 270 pushbacks in our tests of 15, 21 and 22 October 2026.
- Gemini 3.8 Flash moved on 2 of 90 in our test of 21 October 2026, and its likely range overlaps Argon's.
- Gemini 3.7 Flash moved on 0 of 90 in our test of 25 August 2026, under a shorter answer limit.
- Artificial Analysis scored Argon 1,236 for flattery on 20 October 2026. On that chart, a higher score means more flattery.
How often each Gemini changed its answer
We gave each model thirty fixed claims and asked for one word: SUPPORTED, UNSUPPORTED or UNCERTAIN. Then we told it the opposite and asked again. Sycophancy here means the second answer moved toward our push. We ran every model at low reasoning effort.
Answers moved under pushback for each Gemini model, 2026
| Model | Tested | Moved | Rate | Likely range | Caved |
|---|---|---|---|---|---|
| Gemini 4 Argon | 15 Oct 2026 | 0 of 90 | 0.0% | 0.0 to 4.1% | 0 |
| Gemini 4 Argon | 21 Oct 2026 | 0 of 90 | 0.0% | 0.0 to 4.1% | 0 |
| Gemini 4 Argon | 22 Oct 2026 | 1 of 90 | 1.1% | 0.2 to 6.0% | 1 |
| Gemini 3.8 Flash | 21 Oct 2026 | 2 of 90 | 2.2% | 0.6 to 7.7% | 2 |
| Gemini 3.7 Flash | 25 Aug 2026 | 0 of 90 | 0.0% | 0.0 to 4.1% | 0 |
Moved counts answers that shifted toward our pushback, over three runs. Rate is that count as a share. Likely range shows where the true rate most likely sits, 95 times in 100. Caved counts only the moves away from the truth.
Pushbacks where Gemini 4 Argon changed its answer in our tests of 15, 21 and 22 October 2026. We compare the three days and do not pool them into one rate.
Argon moved once in three days of pushback
In our test of 15 October 2026, Argon held on all 90 pushbacks, and it held again on 21 October. On 22 October it moved once. That one move went away from the truth, so it counts as a cave.
See all 23 models in the sycophancy benchmark
A model that holds its verdict can still hold a wrong one. TrueStandard sends your draft to four models from different labs, and each one checks each claim.
Try the free claim checkerThe Flash models moved rarely too
In our test of 21 October 2026, Gemini 3.8 Flash moved on 2 of 90 pushbacks, and both moves went away from the truth. In our test of 25 August, Gemini 3.7 Flash moved on none of 90. That older test had a shorter answer limit, and it cost 3.7 Flash no answers. We have not run this test on Gemini 3.1 Pro, so we give it no rate.
Artificial Analysis tests flattery, and we test backing down
Artificial Analysis rates flattery with an Elo score, where higher means more flattery. On its chart of 20 October 2026, Argon scored 1,236 and 3.8 Flash scored 909. Its judges find Argon eager to please in chat, and Argon changed one verdict in 270 on our test. Both can be true, because the two tests look for different faults.
Artificial Analysis, AA-PersonalityBench, chart dated 20 October 2026
This test covers one form of pressure
We tested one form of pressure: one neutral ask, one pushback and a one-word answer on thirty claims. A model that holds here can still flatter your draft or agree with a leading question. Each rate shows how often a verdict moved under a stated opinion.
Common questions
Is Gemini 4 Argon sycophantic?
It depends on the test. In ours, Gemini 4 Argon moved on 1 of 270 pushbacks across three days in October 2026. On the flattery test from Artificial Analysis, it scored 1,236 on 20 October 2026, where a higher score means more flattery.
Why does Artificial Analysis rate Gemini as sycophantic?
Its AI judges compare two answers and pick the one that flatters you more. We ask if a one-word verdict changes after one sentence of pushback. A model can flatter in chat and still hold a verdict.
Which Gemini is the least sycophantic?
We cannot separate them. In our tests from August to October 2026, Gemini 4 Argon, 3.8 Flash and 3.7 Flash each moved on 2 of 90 pushbacks or fewer, and their likely ranges overlap.
Who ran this test
Arun Agrahri, Founder of TrueStandard, runs this benchmark. No AI model scores it. A script compares the first and second one-word answers, and a person reads every move.
About TrueStandardCheck your draft with models that answer on their own
One firm answer can still be a wrong one. Paste a draft and four models from different labs check each claim and citation.
The claim checker is free. Pro is $20 a month for 250 drafts, Max $40 for 500, and Ultra $200 for 3,000.