‹ PublicAI IndexGoogle
The LLM benchmark aggregator.
Gemini 3.6 Flash
Strongest in Science (#8 of 122), weakest in Safety (#273 of 329). Above par in 19 of 28 scopes. Among the models it meets almost everywhere, it finishes behind Gemini 4 Argon and Gemini 3.7 Flash and ahead of Kimi K2 Thinking and DeepSeek V4 Pro.
§ 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.
Human preference64.6−4.2#21/3541/1
Human preference64.6−4.2#21/3541/1
Core abilities57.5−7.8#25/1991/4
Instruction following63.1−10.7#9/621/1
Language57.7−8.9#15/621/1
Knowledge57.2−5.3#25/1972/3
Academic knowledge60.1−3.8#13/1231/1
Document parsing57−6.3#26/831/1
Reasoning53.2−12.2#63/1953/5
Science62.1−1.4#8/1221/1
Reasoning54.1−14.9#48/1471/3
Mathematics46.3−17.5#101/1521/2
Agents53−13.4#79/2982/6
Knowledge work54.9−15.8#63/2082/2
Professional51.5−10.9#80/1941/2
Finance57.4−5#25/1621/1
Medical56.9−9.6#29/1491/1
Biology research33.7−37#30/311/1
Cybersecurity32.1−31.8#41/471/1
Legal52−13.9#77/1731/1
Coding49.8−16.4#89/1792/6
Agentic coding49.8−17.1#87/1732/5
Safety45.6−15.8#273/3291/3
Secure code52.7−13.5#130/2701/1
Fairness46.2−23.8#160/2921/2
Toxicity avoidance52.3−6.7#165/2681/1
Harm refusal44.9−16.6#241/2921/2
Jailbreak resistance31.3−33.7#248/2681/1
§ 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 3.6 Flash, left for the other.
§ 3 · Sources
Where the numbers come from
8 publications, 39 figures. Every one links to the page it was read from.
LMArena Text 1483
GDPval-AA 1295
AA-Briefcase 956
ARC-AGI-2 60.4%
LiveBench 1.0 73.6LiveBench 1.0 · Reasoning 85.1LiveBench 1.0 · Coding 77.9LiveBench 1.0 · Agentic Coding 43.4LiveBench 1.0 · Mathematics 86.4LiveBench 1.0 · Data Analysis 63LiveBench 1.0 · Language 83.9LiveBench 1.0 · Instruction Following 75.4
Vals · Legal Research Bench 25%Vals · LegalBench 86.7%Vals · Finance Agent 56.3%Vals · CorpFin 63.33%Vals · TaxEval 74.86%Vals · MortgageTax 67.61%Vals · MedCode 53.15%Vals · MedScribe 79.66%Vals · BioMysteryBench 58.52%Vals · CyberBench 45.36%Vals · SWE-bench Verified 79.6% (Mini-SWE-agent)Vals · Vibe Code Bench 64% (OpenHands)Vals · Code Migration 30.93%Vals · GPQA Diamond 93.43%Vals · MMLU Pro 89.28%
ParseBench · Overall 69.12%ParseBench · Tables 89.33%ParseBench · Charts 34.53%ParseBench · Content faithfulness 87.93%ParseBench · Semantic formatting 73.14%ParseBench · Visual grounding 60.65%
Enkrypt · Jailbreak risk 28.5%Enkrypt · Harmful content risk 8.9%Enkrypt · CBRN risk 52.8%Enkrypt · Toxicity risk 4.6%Enkrypt · Bias risk 83.7%Enkrypt · Insecure code risk 27.6%
Badge
[](https://publicai.io/model-index/m/gemini-3-6-flash)