The LLM benchmark aggregator.

GPT OSS 20B

OpenAI · 21B · open weights

Strongest in Secure code (#36 of 274), weakest in Toxicity avoidance (#254 of 272). Above par in 6 of 22 scopes. Among the models it meets almost everywhere, it finishes behind GPT-5 and GPT-5 Mini and ahead of Gemini 1.5 Pro and Mistral Large 3.

§ 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.

Reasoning50.8−16.1#83/1781/4

Mathematics55.3−10.4#45/1401/2

Science45−18.5#91/1221/1

Knowledge40−23.5#118/1381/2

Academic knowledge38−25.9#105/1231/1

Professional43−18.6#143/1681/1

Medical49.7−17.1#82/1401/1

Finance44.1−18.8#117/1511/1

Legal36.3−30.8#142/1511/1

Core abilities45.1−22.3#144/2042/3

General intelligence43.7−26.4#146/2042/3

Safety50.3−11.2#172/3372/3

Secure code62.8−3.7#36/2741/1

Harm refusal56.1−5.5#47/3002/2

Fairness56.2−13.8#57/3002/2

Safe-prompt compliance34.6−25.8#74/821/1

Jailbreak resistance32.9−32#245/2721/1

Toxicity avoidance30.5−28.5#254/2721/1

Agents45.5−22.6#176/2681/5

Knowledge work43.3−30.3#115/1781/2

Human preference49.1−18.4#203/3421/1

§ 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.

Jamba 1.6 Miniwon 12 · lost 3

Jamba 1.6 Largewon 12 · lost 3

GPT-4o Miniwon 16 · lost 5

Command R Pluswon 12 · lost 4

Mistral Smallwon 14 · lost 5

Llama 4 Scout Instructwon 16 · lost 6

Llama 3.3 Instruct Turbo (70B)won 10 · lost 4

Gemini 1.5 Flashwon 12 · lost 5

GPT-4.1 Nanowon 15 · lost 7

Command Awon 13 · lost 7

Mistral Medium 3.5won 9 · lost 6

Mistral Largewon 10 · lost 7

Mistral Large 3won 9 · lost 7

Gemini 1.5 Prowon 10 · lost 8

Claude 3.5 Haikuwon 9 · lost 9

Gemini 2.0 Flashwon 10 · lost 10

DeepSeek V3won 11 · lost 11

DeepSeek R1won 9 · lost 12

Grok 4.3won 9 · lost 12

Kimi K2.6won 9 · lost 12

Grok 3won 8 · lost 11

Llama 4 Maverick Instructwon 9 · lost 13

Gemma 4 31Bwon 7 · lost 11

Ling 3.0 Flashwon 6 · lost 10

Qwen3.6 Pluswon 5 · lost 10

Gemini 3.5 Flash Litewon 5 · lost 10

GPT-4owon 6 · lost 12

Grok 4.1 Fastwon 6 · lost 12

MiniMax M3won 7 · lost 14

Mimo V2.5 Prowon 7 · lost 14

GLM-5.2won 7 · lost 14

GPT-4.1 Miniwon 7 · lost 15

Gemini 2.5 Prowon 7 · lost 15

O3 Miniwon 6 · lost 14

Qwen3.7 Maxwon 6 · lost 15

Gemini 3.5 Flashwon 6 · lost 15

Gemini 3.6 Flashwon 6 · lost 15

Gemini 2.5 Flash Litewon 5 · lost 13

GPT-5.4 Miniwon 4 · lost 11

GPT-5.4 Nanowon 4 · lost 12

Gemini 3.1 Flash Litewon 4 · lost 12

GPT-5 Nanowon 5 · lost 16

O4 Miniwon 5 · lost 16

Gemini 2.5 Flashwon 5 · lost 16

Grok 4won 5 · lost 16

GPT-4.1won 5 · lost 16

Qwen3 235B A22B Instructwon 5 · lost 16

DeepSeek V3.2won 5 · lost 16

DeepSeek V4 Flashwon 5 · lost 16

GLM-5.3 Flashwon 5 · lost 16

Kimi K2 Thinkingwon 4 · lost 13

Claude 3.5 Sonnetwon 4 · lost 13

Mimo V2.6 Prowon 4 · lost 13

Claude Sonnet 4won 5 · lost 17

O1won 4 · lost 14

Claude 3.7 Sonnetwon 4 · lost 15

GLM-4.5won 4 · lost 15

GLM-5.1won 4 · lost 15

Qwen3.8 27Bwon 3 · lost 12

Claude Opus 4won 4 · lost 16

GLM-4.6won 4 · lost 17

GPT-5.6 Lunawon 4 · lost 17

DeepSeek V4 Prowon 4 · lost 17

Claude Haiku 4.5won 4 · lost 18

Claude Opus 4.6won 3 · lost 14

Grok 4.7won 3 · lost 15

GPT-6 Lunawon 3 · lost 15

GPT-6 Solwon 3 · lost 15

Claude Sonnet 4.6won 3 · lost 16

Kimi K2won 3 · lost 18

Kimi K2.5won 3 · lost 18

Qwen3.8 Maxwon 3 · lost 18

Gemini 3.1 Prowon 2 · lost 14

MiniMax M2.7won 2 · lost 17

O3won 2 · lost 19

Inklingwon 2 · lost 19

Grok 4.6won 2 · lost 19

Grok 4.5won 2 · lost 19

GPT-5.4won 2 · lost 19

GPT-5.6 Solwon 2 · lost 19

Muse Spark 1.1won 2 · lost 19

Gemini 3.8 Flashwon 2 · lost 19

Claude Opus 4.7won 2 · lost 19

Grok 4 Fastwon 1 · lost 13

GLM-5won 1 · lost 13

DeepSeek V4.1 Flashwon 1 · lost 13

Muse Spark 1.2won 1 · lost 13

Claude Opus 4.5won 1 · lost 14

GLM-5.3won 1 · lost 14

Gemini 3.7 Flashwon 1 · lost 14

Kimi K3won 1 · lost 14

Gemini 3 Prowon 1 · lost 17

Claude Sonnet 5won 1 · lost 20

GPT-5.6 Terrawon 1 · lost 20

Claude Opus 4.8won 1 · lost 20

GPT-5.5won 1 · lost 20

Claude Opus 5won 1 · lost 20

GPT OSS 120Bwon 1 · lost 21

Claude Sonnet 4.5won 1 · lost 21

Claude Fable 5.1won 0 · lost 15

Claude Fable 5won 0 · lost 15

Claude Opus 5.5won 0 · lost 18

GPT-5.2won 0 · lost 20

GPT-5.1won 0 · lost 20

GPT-5 Miniwon 0 · lost 22

GPT-5won 0 · lost 22

§ 3 · Sources

Where the numbers come from

7 publications, 23 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1318

Artificial Analysis ↗1 measureread 2026-09-26

Artificial Analysis Intelligence Index 9

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 326

Kagi LLM Benchmark ↗1 measureread 2026-09-26

Kagi LLM Benchmark 53.2%

Vals.ai ↗8 measuresread 2026-09-26

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 ↗5 measuresread 2026-09-26

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 AI Safety Leaderboard ↗6 measuresread 2026-09-26

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

[![PublicAI Index](https://publicai.io/model-index/badge?model=gpt-oss-20b)](https://publicai.io/model-index/m/gpt-oss-20b)