BaitBench

openai/gpt-5.4-mini

Corpus v1-2026Q3

Ad-Resistance Score

91.7%

95% CI [90.1–93.2]

Got the honest answer right

98.9%

95% CI [96.7–100.0]

Recommended the wrong product

0.3%

95% CI [0.0–0.9]

Repeated a fake claim as fact

0.0%

Spotted the manipulation

63.5%

95% CI [59.4–67.6]

Over-suspicious of honest info

17.8%

95% CI [10.0–26.7]

Cited its source when echoing a claim

20.4%

Task fails

0

Cost per run

$0.0068

Some of these are scored by an AI judge, not just an automatic check — the methodology page explains exactly how each number is measured.

How often it caught each kind of trick

Bigger shape = it noticed the manipulation more often, across all four trick types.

Fakereviews Puffery Clickbait AI-targeted

By trick type

How it did against each kind of trick.

0 25 50 75 100 Fake reviews Wrong pick 1 Repeated fake claim 0 Spotted it 57 Puffery Wrong pick 0 Repeated fake claim 0 Spotted it 49 Clickbait Wrong pick 1 Repeated fake claim 0 Spotted it 78 AI-targeted Wrong pick 0 Repeated fake claim 0 Spotted it 70
FR FCER_raw FCER MDR
A — fake reviews 0.6% 0.0% 0.0% 56.7%
B — puffery 0.0% 2.2% 0.0% 48.9%
C — clickbait 0.6% 0.0% 0.0% 78.3%
D — AI injection 0.0% 1.1% 0.0% 70.0%

By how obvious the trick was

Subtle fakes vs blatant, over-the-top ones.

0 25 50 75 100 Subtle Wrong pick 0 Repeated fake claim 0 Spotted it 49 Blatant Wrong pick 1 Repeated fake claim 0 Spotted it 78
FR FCER_raw FCER MDR
L1 — subtle 0.0% 0.7% 0.0% 48.6%
L3 — blatant 0.6% 1.0% 0.0% 78.3%

Judge audit

Judge model: google/gemini-2.5-flash · Prompt version: j2

MDR agreement: 239/240 (99.6%)

Attribution agreement: 2/2 (100.0%)

Cost

Total: $1.8383 · Per run: $0.0068

← Back to leaderboard