‹ PublicAI Index
The LLM benchmark aggregator.
Llama 3.1 405B Instruct
Meta · 405B
Strongest in Safe-prompt compliance (#41 of 82), weakest in Secure code (#233 of 268). Above par in 8 of 14 scopes. Among the models it meets almost everywhere, it finishes behind Claude Sonnet 4.5 and GPT-5 and ahead of GPT OSS 20B and Gemini 2.0 Flash-Lite.
§ 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.
Professional48.5−13.9#116/1941/2
Medical52.5−14#68/1491/1
Finance43.6−18.8#130/1621/1
Reasoning42.2−23.2#160/1951/5
Reasoning37.5−31.5#132/1471/3
Safety50.1−11.3#175/3272/3
Safe-prompt compliance53.9−6.5#41/821/1
Jailbreak resistance59−6#59/2661/1
Fairness49.4−20.6#119/2902/2
Harm refusal50.6−10.9#142/2902/2
Toxicity avoidance53.1−5.9#146/2661/1
Secure code33.2−33#233/2681/1
Human preference50.4−18.4#195/3541/1
Human preference50.4−18.4#195/3541/1
§ 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 Llama 3.1 405B Instruct, left for the other.
§ 3 · Sources
Where the numbers come from
5 publications, 15 figures. Every one links to the page it was read from.
LMArena Text 1335
SimpleBench 23%
Vals · TaxEval 60.88%Vals · MedQA 88.24%
HELM Safety · HarmBench 62.7%HELM Safety · SimpleSafetyTests 98.8%HELM Safety · Anthropic Red Team 96.5%HELM Safety · BBQ 94.5%HELM Safety · XSTest 95.9%
Enkrypt · Jailbreak risk 5.3%Enkrypt · Harmful content risk 26.7%Enkrypt · CBRN risk 9.8%Enkrypt · Toxicity risk 4%Enkrypt · Bias risk 87.1%Enkrypt · Insecure code risk 67.6%
Badge
[](https://publicai.io/model-index/m/llama-3-1-405b-instruct)