‹ PublicAI IndexGoogle
The LLM benchmark aggregator.
Gemma 4 E4B
Strongest in Secure code (#27 of 270), weakest in Fairness (#272 of 292). Above par in 2 of 15 scopes. Among the models it meets almost everywhere, it finishes behind MiniMax M2.7 and Claude Opus 5.5 and ahead of MiniMax M3 and GLM-5.2.
§ 1 · Profile
What it is good at
Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.
Knowledge39.8−22.7#184/1971/3
Document parsing39.1−24.2#69/831/1
Safety48.3−13.1#223/3291/3
Secure code63.2−3#27/2701/1
Toxicity avoidance56.1−2.9#70/2681/1
Harm refusal47.1−14.4#202/2921/2
Jailbreak resistance38.3−26.7#222/2681/1
Fairness39.7−30.3#272/2921/2
Agents42.9−23.5#253/2981/6
Knowledge work35.8−34.9#178/2081/2
Tool use37.8−36.2—/81✱0/1
Reasoning47.5−17.9—/195✱0/5
Science42.2−21.3—/122✱0/1
Coding45.6−20.6—/179✱0/6
Agentic coding45−21.9—/173✱0/5
§ 2 · Head to head
What it beats, and what beats it
The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Gemma 4 E4B, left for the other.
§ 3 · Sources
Where the numbers come from
4 publications, 16 figures. Every one links to the page it was read from.
GDPval-AA -29
ParseBench · Overall 41.13%ParseBench · Tables 43.05%ParseBench · Charts 5.86%ParseBench · Content faithfulness 83.9%ParseBench · Semantic formatting 41.03%ParseBench · Visual grounding 31.79%
Enkrypt · Jailbreak risk 22.6%Enkrypt · Harmful content risk 14.4%Enkrypt · CBRN risk 33.2%Enkrypt · Toxicity risk 2%Enkrypt · Bias risk 93.8%Enkrypt · Insecure code risk 6.2%
tau3-Banking 5%Terminal-Bench 2.1 2%GPQA Diamond 52%
Badge
[](https://publicai.io/model-index/m/gemma-4-e4b)