‹ PublicAI Index
The LLM benchmark aggregator.
GPT OSS 20B
OpenAI · 21B · open weights
Strongest in Secure code (#37 of 270), weakest in Toxicity avoidance (#251 of 268). Above par in 8 of 22 scopes. Among the models it meets almost everywhere, it finishes behind GPT-5 and GPT-5 mini and ahead of Gemma 4 31B and Gemini 1.5 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.
Reasoning50.8−14.6#91/1951/5
Mathematics55.3−8.5#51/1521/2
Science45−18.5#91/1221/1
Core abilities50.2−15.1#96/1991/4
General intelligence50.2−19.2#85/1781/3
Knowledge45−17.5#152/1971/3
Academic knowledge38−25.9#105/1231/1
Professional43.2−19.2#165/1941/2
Medical49.7−16.8#90/1491/1
Finance44.1−18.3#127/1621/1
Legal36.3−29.6#163/1731/1
Safety50.3−11.1#171/3292/3
Secure code62.5−3.7#37/2701/1
Harm refusal56.2−5.3#45/2922/2
Fairness56.2−13.8#56/2922/2
Safe-prompt compliance34.6−25.8#74/821/1
Jailbreak resistance32.8−32.2#242/2681/1
Toxicity avoidance29.8−29.2#251/2681/1
Agents46−20.4#199/2981/6
Knowledge work42−28.7#139/2081/2
Human preference48.8−20#212/3541/1
Human preference48.8−20#212/3541/1
§ 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 GPT OSS 20B, left for the other.
§ 3 · Sources
Where the numbers come from
6 publications, 22 figures. Every one links to the page it was read from.
LMArena Text 1318
GDPval-AA 347
Kagi LLM Benchmark 53.2%
Vals · CaseLaw 43.84%Vals · LegalBench 70.85%Vals · CorpFin 53.15%Vals · TaxEval 63.7%Vals · MedQA 82.88%Vals · GPQA Diamond 68.94%Vals · MMLU Pro 71.64%Vals · AIME 86.04%
HELM Safety · HarmBench 98.7%HELM Safety · SimpleSafetyTests 100%HELM Safety · Anthropic Red Team 99.7%HELM Safety · BBQ 96.7%HELM Safety · XSTest 87.2%
Enkrypt · Jailbreak risk 27.2%Enkrypt · Harmful content risk 19.4%Enkrypt · CBRN risk 19%Enkrypt · Toxicity risk 20%Enkrypt · Bias risk 72.4%Enkrypt · Insecure code risk 7.6%
Badge
[](https://publicai.io/model-index/m/gpt-oss-20b)