‹ PublicAI IndexGoogle
The LLM benchmark aggregator.
Gemini 2.5 Pro
Strongest in Code generation (#3 of 20), weakest in Jailbreak resistance (#255 of 268). Above par in 19 of 27 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of O3 and Qwen3.7 Max.
§ 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.
Professional54.7−7.7#32/1941/2
Legal56.5−9.4#29/1731/1
Finance55.1−7.3#47/1621/1
Medical54.2−12.3#51/1491/1
Core abilities56−9.3#34/1991/4
General intelligence58.6−10.8#31/1781/3
Human preference61.1−7.7#72/3541/1
Human preference61.1−7.7#72/3541/1
Reasoning52.1−13.3#80/1953/5
Mathematics55.2−8.6#52/1521/2
Science53.3−10.2#66/1221/1
Reasoning49.1−19.9#71/1472/3
Coding49.1−17.1#94/1793/6
Code generation61.1−4.2#3/201/1
Agentic coding45.8−21.1#108/1732/5
Knowledge51.5−11#95/1971/3
Academic knowledge53.5−10.4#65/1231/1
Safety50.2−11.2#175/3293/3
Safe-prompt compliance60.2−0.2#4/821/1
Factual grounding57.5−13.2#32/1041/1
Fairness60.4−9.6#40/2922/2
Toxicity avoidance53.2−5.8#146/2681/1
Secure code47.4−18.8#181/2701/1
Harm refusal47−14.5#203/2922/2
Jailbreak resistance28.2−36.8#255/2681/1
Agents43−23.4#251/2982/6
Knowledge work38.7−32#160/2082/2
§ 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 Gemini 2.5 Pro, left for the other.
§ 3 · Sources
Where the numbers come from
12 publications, 33 figures. Every one links to the page it was read from.
LMArena Text 1446
GDPval-AA 463
AA-Briefcase 303
ARC-AGI-2 4.9%
LiveCodeBench 73.6%
Aider polyglot 83.1%
Kagi LLM Benchmark 70.3%
SimpleBench 62.4%
Vals · CaseLaw 63.88%Vals · LegalBench 84.32%Vals · CorpFin 60.8%Vals · TaxEval 72.89%Vals · MortgageTax 68.92%Vals · MedQA 93.14%Vals · MedCode 50.59%Vals · MedScribe 73.55%Vals · SWE-bench Verified 54.4% (Mini-SWE-agent)Vals · Vibe Code Bench 0.4% (OpenHands)Vals · GPQA Diamond 80.81%Vals · MMLU Pro 84.06%Vals · AIME 85.83%
HELM Safety · HarmBench 65.4%HELM Safety · SimpleSafetyTests 97%HELM Safety · Anthropic Red Team 99.5%HELM Safety · BBQ 96.4%HELM Safety · XSTest 98.7%
Enkrypt · Jailbreak risk 31.1%Enkrypt · Harmful content risk 54.6%Enkrypt · CBRN risk 20.8%Enkrypt · Toxicity risk 4%Enkrypt · Bias risk 61.2%Enkrypt · Insecure code risk 38.5%
Vectara · Factual consistency 93%
Badge
[](https://publicai.io/model-index/m/gemini-2-5-pro)