‹ PublicAI Index
The LLM benchmark aggregator.
Gemma 3 27B IT
Google · 27B · open weights
Strongest in Factual grounding (#35 of 104), weakest in Agents (#280 of 298). Above par in 6 of 16 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 4.8 and Grok 4.7 and ahead of Mercury 2 and Amazon Nova Micro.
§ 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.
Core abilities45.8−19.5#137/1991/4
General intelligence43.9−25.5#127/1781/3
Safety51.6−9.8#141/3292/3
Factual grounding56.4−14.3#35/1041/1
Jailbreak resistance56.9−8.1#82/2681/1
Toxicity avoidance55.2−3.8#95/2681/1
Fairness46.7−23.3#153/2921/2
Harm refusal49.8−11.7#157/2921/2
Secure code48.6−17.6#169/2701/1
Coding41.7−24.5#159/1791/6
Agentic coding40.9−26#156/1731/5
Human preference53.3−15.5#162/3541/1
Human preference53.3−15.5#162/3541/1
Agents40.6−25.8#280/2982/6
Tool use43.9−30.1#49/811/1
Knowledge work30−40.7#205/2081/2
§ 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 3 27B IT, left for the other.
§ 3 · Sources
Where the numbers come from
7 publications, 12 figures. Every one links to the page it was read from.
LMArena Text 1365
GDPval-AA -401
Aider polyglot 4.9%
BFCL v4 29.47%
Kagi LLM Benchmark 40.4%
Enkrypt · Jailbreak risk 7%Enkrypt · Harmful content risk 43.9%Enkrypt · CBRN risk 9.7%Enkrypt · Toxicity risk 2.6%Enkrypt · Bias risk 83%Enkrypt · Insecure code risk 36%
Vectara · Factual consistency 92.6%
Badge
[](https://publicai.io/model-index/m/gemma-3-27b-it)