The LLM benchmark aggregator.

GPT OSS 120B

OpenAI · 117B · open weights

Strongest in Secure code (#7 of 270), weakest in Toxicity avoidance (#234 of 268). Above par in 10 of 26 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5.5 and Claude Fable 5 and ahead of Llama 4 Maverick Instruct and Claude Haiku 4.5.

§ 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.

Safety54.2−7.2#70/3293/3

Secure code65.6−0.6#7/2701/1

Harm refusal57.8−3.7#25/2922/2

Fairness61.9−8.1#31/2922/2

Safe-prompt compliance46.8−13.6#63/821/1

Factual grounding39.2−31.5#90/1041/1

Jailbreak resistance42.7−22.3#206/2681/1

Toxicity avoidance43.1−15.9#234/2681/1

Reasoning48.5−16.9#109/1952/5

Mathematics57−6.8#32/1521/2

Science51.7−11.8#73/1221/1

Reasoning37.1−31.9#135/1471/3

Professional48.3−14.1#121/1941/2

Medical54.2−12.3#52/1491/1

Finance51−11.4#91/1621/1

Legal41.6−24.3#147/1731/1

Knowledge48.9−13.6#129/1971/3

Academic knowledge47.4−16.5#91/1231/1

Core abilities45.8−19.5#136/1992/4

General intelligence44.6−24.8#125/1782/3

Coding42−24.2#155/1792/6

Agentic coding41.5−25.4#150/1732/5

Agents48.2−18.2#163/2981/6

Knowledge work46.4−24.3#120/2081/2

Human preference52.1−16.7#173/3541/1

§ 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 120B, left for the other.

Magistral Small 1.2won 16 · lost 1

Gemini 1.5 Flashwon 15 · lost 2

Mistral Medium 3.5won 15 · lost 2

GPT-4o miniwon 21 · lost 3

GPT OSS 20Bwon 19 · lost 3

Llama 3.3 70B Instructwon 16 · lost 4

GPT-4.1 nanowon 19 · lost 6

Llama 4 Scout Instructwon 18 · lost 6

Gemini 1.5 Prowon 14 · lost 5

Command R+won 13 · lost 5

Gemma 4 31Bwon 15 · lost 6

Command Awon 16 · lost 7

Gemini 2.0 Flashwon 15 · lost 8

GPT-4owon 15 · lost 9

Claude 3.5 Haikuwon 12 · lost 8

Mistral Large 3won 12 · lost 8

Gemini 2.5 Flash-Litewon 11 · lost 8

O3 Miniwon 13 · lost 10

Ling 3.0 Flashwon 10 · lost 8

GPT-5 nanowon 12 · lost 10

Grok 4.1 Fastwon 12 · lost 10

Claude Haiku 4.5won 13 · lost 11

Llama 4 Maverick Instructwon 14 · lost 12

DeepSeek V3.1won 8 · lost 8

GPT-5.4 nanowon 10 · lost 10

Grok 4.3won 11 · lost 11

Kimi K2won 12 · lost 13

MiMo V2.5 Prowon 10 · lost 11

Gemini 3.1 Flash-Litewon 9 · lost 10

DeepSeek V3won 12 · lost 14

GPT-4.1 miniwon 11 · lost 14

DeepSeek R1won 11 · lost 14

Mistral Large 4won 9 · lost 12

Grok 3won 10 · lost 14

NVIDIA Nemotron 3 Ultra 550B A55Bwon 10 · lost 14

Grok 4 Fastwon 7 · lost 11

GLM-4.5won 7 · lost 12

GPT-5.4 miniwon 7 · lost 12

O4 Miniwon 9 · lost 16

Gemini 2.5 Flashwon 9 · lost 16

Qwen3 235B A22B Instructwon 9 · lost 16

Claude 3.5 Sonnetwon 7 · lost 13

MiMo V2.6 Flashwon 7 · lost 13

Grok 4won 8 · lost 16

GLM-4.6won 8 · lost 16

MiniMax M3won 8 · lost 16

Inklingwon 8 · lost 16

DeepSeek V3.2won 8 · lost 17

Claude 3.7 Sonnetwon 7 · lost 15

GPT-6 Lunawon 7 · lost 15

MiniMax M2.1won 5 · lost 11

Grok 4.20won 5 · lost 11

Gemini 2.5 Prowon 8 · lost 18

Kimi K2.6won 7 · lost 16

Qwen3.7 Maxwon 7 · lost 16

DeepSeek V4 Flashwon 7 · lost 17

Kimi K2.5won 7 · lost 17

GPT-5.6 Lunawon 7 · lost 17

GLM-5.3 Flashwon 7 · lost 17

GLM-5.2won 7 · lost 17

MiMo V2.6 Prowon 6 · lost 15

GPT-4.1won 7 · lost 18

Inkling Smallwon 5 · lost 13

Qwen3.8 27Bwon 5 · lost 13

Gemini 3.5 Flash-Litewon 5 · lost 13

GLM-5.1won 6 · lost 16

Claude Sonnet 4won 7 · lost 19

Gemini 3.5 Flashwon 6 · lost 17

Gemini 3.6 Flashwon 6 · lost 17

Kimi K2 Thinkingwon 5 · lost 15

Grok 3 Miniwon 5 · lost 15

Claude Opus 4won 6 · lost 18

Qwen3.8 Maxwon 6 · lost 18

DeepSeek V4 Prowon 6 · lost 19

O1won 5 · lost 16

GPT-5 miniwon 6 · lost 20

MiniMax M2.7won 5 · lost 17

GPT-6 Solwon 5 · lost 17

Claude Sonnet 4.6won 5 · lost 17

GLM-4.7won 4 · lost 14

Gemini 3.1 Prowon 4 · lost 15

Gemini 3.8 Flashwon 5 · lost 19

GPT-5.6 Solwon 5 · lost 20

Claude Opus 4.6won 4 · lost 17

Qwen3.6 Pluswon 3 · lost 13

MiniMax M2.5won 3 · lost 15

GPT-5.6 Terrawon 4 · lost 20

GPT-5.4won 4 · lost 21

Grok 4.7won 3 · lost 18

Gemini 3 Prowon 3 · lost 19

Claude Sonnet 5won 3 · lost 20

O3won 3 · lost 21

Grok 4.5won 3 · lost 21

GPT-5.5won 3 · lost 22

Claude Sonnet 4.5won 3 · lost 23

GLM-5won 2 · lost 16

DeepSeek V4.1 Flashwon 2 · lost 16

Claude Opus 4.5won 2 · lost 17

Muse Spark 1.1won 2 · lost 20

GPT-5.1won 2 · lost 22

Grok 4.6won 2 · lost 22

Claude Opus 4.8won 2 · lost 22

Claude Opus 4.7won 2 · lost 23

GPT-5won 2 · lost 24

Claude Haiku 5.5won 1 · lost 15

Gemini 3 Flashwon 1 · lost 15

Muse Spark 1.2won 1 · lost 15

Gemini 3.7 Flashwon 1 · lost 16

Gemini 4 Argonwon 1 · lost 16

GPT-6 Astrawon 1 · lost 17

GLM-5.3won 1 · lost 17

Kimi K3won 1 · lost 17

Claude Opus 5won 1 · lost 22

GPT-5.2won 1 · lost 23

Claude Sonnet 5.5won 0 · lost 16

GPT-6.1 Solwon 0 · lost 16

Claude Fable 5.1won 0 · lost 18

Claude Fable 5won 0 · lost 18

Claude Opus 5.5won 0 · lost 23

§ 3 · Sources

Where the numbers come from

10 publications, 27 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-11-10

LMArena Text 1352

Artificial Analysis ↗1 measureread 2026-11-10

Artificial Analysis Intelligence Index 4

GDPval-AA ↗1 measureread 2026-11-10

GDPval-AA 616

Aider polyglot ↗1 measureread 2026-11-10

Aider polyglot 41.8%

Kagi LLM Benchmark ↗1 measureread 2026-11-10

Kagi LLM Benchmark 58.6%

SimpleBench ↗1 measureread 2026-11-10

SimpleBench 22.1%

Vals.ai ↗9 measuresread 2026-11-10

Vals · CaseLaw 48.77%Vals · LegalBench 75.94%Vals · CorpFin 58.24%Vals · TaxEval 71.59%Vals · MedQA 91.36%Vals · SWE-bench Verified 33.6% (Mini-SWE-agent)Vals · GPQA Diamond 78.54%Vals · MMLU Pro 79.17%Vals · AIME 92.6%

HELM Safety ↗5 measuresread 2026-11-10

HELM Safety · HarmBench 100%HELM Safety · SimpleSafetyTests 100%HELM Safety · Anthropic Red Team 99.5%HELM Safety · BBQ 98.5%HELM Safety · XSTest 92.7%

Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-11-10

Enkrypt · Jailbreak risk 18.9%Enkrypt · Harmful content risk 14.4%Enkrypt · CBRN risk 13.8%Enkrypt · Toxicity risk 10.9%Enkrypt · Bias risk 60.2%Enkrypt · Insecure code risk 1.3%

Vectara Hallucination Leaderboard ↗1 measureread 2026-11-10

Vectara · Factual consistency 85.8%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=gpt-oss-120b)](https://publicai.io/model-index/m/gpt-oss-120b)