Claude Sycophancy Rate by Version
By Arun Agrahri, Founder of TrueStandard. Published .
In our test, Claude Sonnet 5.5 changed its answer on 8 of 90 pushbacks on 21 October 2026 and on 9 of 90 the next day. Claude Opus 5.5 moved on 0, then 1. Claude Fable 5.1 moved on 1, then 3. All three ran the same thirty claims on those two days, and Sonnet 5.5 moved more often than any other Claude we tested.
Key takeaways
- Claude Opus 5.5 moved on 0 of 90 pushbacks in our test of 21 October 2026, and 1 of 90 on 22 October.
- Claude Sonnet 5.5 moved on 8 of 90, then 9 of 90, in the same test on 21 and 22 October 2026.
- Claude Haiku 5.5, the fast tier, moved on 1 of 90 in our test of 21 October 2026.
- Claude Sonnet 5 moved on 3 of 90 in our test of 25 August 2026. That test had a shorter answer limit, so we do not call Sonnet 5.5's rate a decline.
- Claude Fable 5.1 did well on our pushback test of October 2026 and on the flattery test from Artificial Analysis.
How often each Claude 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 Claude model, 2026
| Model | Tested | Moved | Rate | Likely range | Caved |
|---|---|---|---|---|---|
| Claude Opus 5.5 | 21 Oct 2026 | 0 of 90 | 0.0% | 0.0 to 4.1% | 0 |
| Claude Opus 5.5 | 22 Oct 2026 | 1 of 90 | 1.1% | 0.2 to 6.0% | 1 |
| Claude Fable 5.1 | 21 Oct 2026 | 1 of 89 | 1.1% | 0.2 to 6.1% | 1 |
| Claude Fable 5.1 | 22 Oct 2026 | 3 of 90 | 3.3% | 1.1 to 9.3% | 3 |
| Claude Haiku 5.5 | 21 Oct 2026 | 1 of 90 | 1.1% | 0.2 to 6.0% | 1 |
| Claude Sonnet 5.5 | 21 Oct 2026 | 8 of 90 | 8.9% | 4.6 to 16.6% | 8 |
| Claude Sonnet 5.5 | 22 Oct 2026 | 9 of 90 | 10.0% | 5.4 to 17.9% | 6 |
| Claude Sonnet 5 | 25 Aug 2026 | 3 of 90 | 3.3% | 1.1 to 9.3% | 1 |
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 Claude Sonnet 5.5 changed its answer in our test of 22 October 2026. Claude Opus 5.5 moved on 1 of 90 in the same test.
Sonnet 5.5 moved on about one pushback in ten
In our test of 21 October 2026, all 8 of Sonnet 5.5's moves went away from the truth. On 22 October, 6 of its 9 did, and the other 3 moved toward the truth. Opus 5.5 and Sonnet 5.5 sit in different price tiers, so compare each one within its own tier.
See all 23 models in the sycophancy benchmark
Tell a model it is wrong and it may give way, even when it was right. Asking the same model again will not show you that. TrueStandard sends your draft to four models from different labs, and each one checks each claim.
Try the free claim checkerOpus 5.5, Haiku 5.5 and Fable 5.1 moved rarely
In our test of 21 October 2026, Opus 5.5 moved on none of its 90 pushbacks, and Haiku 5.5 and Fable 5.1 moved once each. On 22 October, Opus 5.5 moved once and Fable 5.1 moved 3 times. Their likely ranges overlap, so we do not rank these three.
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, Fable 5.1 scored 723, Sonnet 5.5 907, Opus 5.5 991 and Haiku 5.5 1,004. It ran Opus 5.5 and Fable 5.1 at maximum reasoning, and we ran every model at low. So Sonnet 5.5 can flatter less on that test and back down more on ours.
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 Opus 5.5 sycophantic?
Rarely in our pushback test. Claude Opus 5.5 changed 0 of 90 answers on 21 October 2026 and 1 of 90 on 22 October. On the flattery test from Artificial Analysis, its Elo score was 991 on 20 October 2026.
Is Fable 5.1 sycophantic?
It did well on both tests we can compare. In ours, Claude Fable 5.1 moved on 1 of 89 pushbacks on 21 October 2026 and 3 of 90 on 22 October. Artificial Analysis rated it second least sycophantic of 38 models on 20 October 2026.
Which Claude is the least sycophantic?
In our pushback test, Claude Opus 5.5 moved least: 0 and 1 of 90 on 21 and 22 October 2026. Its likely range overlaps Haiku 5.5 and Fable 5.1, so we cannot separate those three. Sonnet 5.5 moved on 8 and 9 of 90.
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
A model can drop a right answer when you push it. 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.