‹ PublicAI IndexGoogle
The LLM benchmark aggregator.
Gemini 3.5 Flash
Strongest in Instruction following (#7 of 62), weakest in Harm refusal (#257 of 292). Above par in 18 of 27 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5.1 and Claude Opus 5.5 and ahead of GPT-5.6 Terra and Mistral Large 4.
§ 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.
Knowledge59.3−3.2#11/1972/3
Academic knowledge60.4−3.5#10/1231/1
Document parsing59.5−3.8#14/831/1
Core abilities58−7.3#22/1991/4
Instruction following63.5−10.3#7/621/1
Language58.6−8#14/621/1
Professional55.7−6.7#26/1941/2
Finance58.2−4.2#13/1621/1
Medical57.1−9.4#27/1491/1
Legal52.2−13.7#75/1731/1
Human preference64.2−4.6#28/3541/1
Human preference64.2−4.6#28/3541/1
Reasoning53.9−11.5#58/1954/5
Science61.6−1.9#15/1221/1
Reasoning61.2−7.8#17/1472/3
Mathematics44.4−19.4#110/1522/2
Coding49.3−16.9#92/1792/6
Agentic coding49.3−17.6#90/1732/5
Agents49.3−17.1#142/2983/6
Web research41.9−22#28/351/2
Knowledge work53.1−17.6#75/2082/2
Safety46.7−14.7#256/3291/3
Secure code60.5−5.7#60/2701/1
Toxicity avoidance55.1−3.9#102/2681/1
Fairness48.4−21.6#134/2921/2
Jailbreak resistance28.8−36.2#252/2681/1
Harm refusal43.7−17.8#257/2921/2
§ 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 3.5 Flash, left for the other.
§ 3 · Sources
Where the numbers come from
9 publications, 40 figures. Every one links to the page it was read from.
LMArena Text 1478
GDPval-AA 1212
AA-Briefcase 876
ARC-AGI-2 72.1%
LiveBench 1.0 74.6LiveBench 1.0 · Reasoning 82LiveBench 1.0 · Coding 78.2LiveBench 1.0 · Agentic Coding 49LiveBench 1.0 · Mathematics 88.2LiveBench 1.0 · Data Analysis 64.9LiveBench 1.0 · Language 84.6LiveBench 1.0 · Instruction Following 75.6
SimpleBench 76.7%
Vals · Legal Research Bench 30.77%Vals · LegalBench 83.6%Vals · Finance Agent 57.86%Vals · CorpFin 64.69%Vals · TaxEval 74.37%Vals · MortgageTax 68.12%Vals · MedCode 55.83%Vals · MedScribe 76.57%Vals · Web Search Index 41.36% (Exa)Vals · SWE-bench Verified 78.8% (Mini-SWE-agent)Vals · Vibe Code Bench 48.68% (OpenHands)Vals · Code Migration 26.75%Vals · GPQA Diamond 92.68%Vals · MMLU Pro 89.52%Vals · ProofBench 31%
ParseBench · Overall 73.22%ParseBench · Tables 90.48%ParseBench · Charts 45.27%ParseBench · Content faithfulness 88.52%ParseBench · Semantic formatting 77.87%ParseBench · Visual grounding 63.95%
Enkrypt · Jailbreak risk 30.6%Enkrypt · Harmful content risk 15%Enkrypt · CBRN risk 52%Enkrypt · Toxicity risk 2.7%Enkrypt · Bias risk 80.4%Enkrypt · Insecure code risk 11.6%
Badge
[](https://publicai.io/model-index/m/gemini-3-5-flash)