The LLM benchmark aggregator.

Gemma 3 4B IT

Google · 4.3B · open weights

Strongest in Factual grounding (#28 of 104), weakest in Human preference (#230 of 354). Above par in 3 of 13 scopes. Among the models it meets almost everywhere, it finishes behind GPT-4.1 mini and Claude Opus 4.8 and ahead of Grok 4.1 Fast and GPT-4.

§ 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.

Core abilities40.6−24.7#184/1991/4

General intelligence36.5−32.9#169/1781/3

Safety49.6−11.8#188/3292/3

Factual grounding59−11.7#28/1041/1

Jailbreak resistance52.3−12.7#154/2681/1

Fairness46−24#168/2921/2

Toxicity avoidance51.7−7.3#180/2681/1

Harm refusal47.7−13.8#192/2921/2

Secure code44.7−21.5#195/2701/1

Agents44.5−21.9#229/2981/6

Tool use36.8−37.2#74/811/1

Human preference47.4−21.4#230/3541/1

§ 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 Gemma 3 4B IT, left for the other.

Magistral Smallwon 8 · lost 0

SmolLM2 1.7B Instructwon 8 · lost 0

Zephyr 7B Betawon 8 · lost 0

Nous Hermes 2 Mixtral 8x7B DPOwon 8 · lost 0

Jamba Mini 1.7won 8 · lost 1

Command R+won 8 · lost 1

OLMo 7B Instructwon 7 · lost 1

Gemma 2B ITwon 7 · lost 1

Gemma 7B ITwon 7 · lost 1

DeepSeek LLM 67B Chatwon 7 · lost 1

DBRX Instructwon 7 · lost 1

Mixtral 8x7B Instructwon 7 · lost 1

Jamba 1.5 Miniwon 7 · lost 1

Jamba 1.5 Largewon 7 · lost 1

Mistral 7B Instruct v0.2won 6 · lost 2

GPT-3.5 Turbowon 6 · lost 2

Magistral Mediumwon 7 · lost 3

Phi-4 Mini Instructwon 5 · lost 3

Grok Build 0.1won 5 · lost 3

Granite 3.1 2B Instructwon 5 · lost 3

Qwen1.5 14B Chatwon 5 · lost 3

Gemma 2 2B ITwon 5 · lost 3

Llama 3.1 70B Instructwon 5 · lost 3

Llama 3.2 1B Instructwon 6 · lost 4

Jamba Large 1.7won 5 · lost 4

Mistral Small 3 24B Instructwon 5 · lost 4

GPT-4won 5 · lost 4

Grok 4.1 Fastwon 6 · lost 5

Command R7Bwon 4 · lost 4

Kimi K2.7 Codewon 4 · lost 4

Starling LM 7B Betawon 4 · lost 4

Mixtral 8x22B Instructwon 4 · lost 4

Llama 3.2 3B Instructwon 5 · lost 5

DeepSeek R1won 6 · lost 6

Grok 3won 5 · lost 6

DeepSeek V4 Flashwon 5 · lost 6

Qwen3 8Bwon 4 · lost 5

Aya Expanse 8Bwon 4 · lost 5

Aya Expanse 32Bwon 4 · lost 5

NVIDIA Nemotron 3 Super 120B A12Bwon 4 · lost 5

DeepSeek V3won 5 · lost 7

Gemma 4 31Bwon 5 · lost 7

Granite 3.1 8B Instructwon 4 · lost 6

Llama 3.1 8B Instructwon 4 · lost 6

Mercury 2won 4 · lost 6

Qwen3 30B A3B Instructwon 4 · lost 6

Gemma 4 26B A4Bwon 4 · lost 6

Grok 4.3won 4 · lost 6

Command Awon 5 · lost 8

GLM-4.6won 5 · lost 8

Granite 3.2 8B Instructwon 3 · lost 5

Phi-3 Mini 128K Instructwon 3 · lost 5

Llama 2 13B Chatwon 3 · lost 5

Phi-3 Mini 4K Instructwon 3 · lost 5

Granite 3.0 2B Instructwon 3 · lost 5

Llama 2 70B Chatwon 3 · lost 5

Phi-3 Small 8K Instructwon 3 · lost 5

Granite 3.0 8B Instructwon 3 · lost 5

Phi-3 Medium 4K Instructwon 3 · lost 5

Llama 3.1 Tulu 3 8Bwon 3 · lost 5

Llama 3 8B Instructwon 3 · lost 5

Llama 3.1 Nemotron 70B Instructwon 3 · lost 5

QwQ 32Bwon 3 · lost 5

Gemini 2.0 Flash-Litewon 3 · lost 5

Amazon Nova Prowon 4 · lost 7

Llama 3.3 70B Instructwon 4 · lost 7

DeepSeek V3.1won 4 · lost 7

NVIDIA Nemotron 3 Ultra 550B A55Bwon 4 · lost 7

MiMo V2.6 Flashwon 4 · lost 7

Kimi K2.6won 4 · lost 7

GLM-5.2won 4 · lost 7

Amazon Nova Litewon 3 · lost 6

MiMo V2.5 Prowon 3 · lost 6

Gemini 2.5 Prowon 4 · lost 8

Llama 4 Scout Instructwon 4 · lost 9

DeepSeek V3.2won 4 · lost 9

Claude 3 Haikuwon 3 · lost 7

Llama 3 70B Instructwon 3 · lost 7

GLM-4.5won 3 · lost 7

Qwen3.7 Maxwon 3 · lost 7

Gemini 3.5 Flashwon 3 · lost 7

Gemini 3.6 Flashwon 3 · lost 7

Amazon Nova Microwon 3 · lost 8

Mistral Large 4won 3 · lost 8

MiniMax M3won 3 · lost 8

GPT-6 Solwon 3 · lost 8

Llama 2 7B Chatwon 2 · lost 6

Qwen2 72B Instructwon 2 · lost 6

Claude 3 Sonnetwon 2 · lost 6

Gemma 2 27B ITwon 2 · lost 6

Qwen2.5 72B Instructwon 2 · lost 6

Qwen3 VL 235B A22B Instructwon 2 · lost 6

Claude Opus 4.5won 2 · lost 6

GPT OSS 120Bwon 3 · lost 9

Grok 4won 3 · lost 9

Kimi K2.5won 3 · lost 9

DeepSeek V4 Prowon 3 · lost 9

Llama 4 Maverick Instructwon 3 · lost 10

Gemini 2.5 Flashwon 3 · lost 10

Kimi K2won 3 · lost 10

Granite 4.0 H Smallwon 2 · lost 7

Qwen3 Next 80B A3B Thinkingwon 2 · lost 7

Qwen3.5 27Bwon 2 · lost 7

Claude Opus 4.6won 2 · lost 7

GPT-4owon 2 · lost 8

Qwen3.5 35B A3Bwon 2 · lost 8

Qwen3.5 122B A10Bwon 2 · lost 8

Phi 4won 2 · lost 9

GPT OSS 20Bwon 2 · lost 9

Gemma 3 12B ITwon 2 · lost 9

GPT-6 Lunawon 2 · lost 9

GLM-5.3 Flashwon 2 · lost 9

GPT-4.1 nanowon 2 · lost 10

GPT-5.4won 2 · lost 10

GPT-5.5won 2 · lost 10

GPT-5.6 Solwon 2 · lost 10

Gemma 3 27B ITwon 2 · lost 11

O4 Miniwon 2 · lost 11

Qwen3 235B A22B Instructwon 2 · lost 11

Claude Sonnet 4.5won 2 · lost 11

Nova Premierwon 1 · lost 7

Kimi K2 Thinkingwon 1 · lost 7

Gemini 1.5 Flashwon 1 · lost 7

Llama 3.1 405B Instructwon 1 · lost 7

Gemini 1.5 Prowon 1 · lost 7

GLM-4.5 Airwon 1 · lost 7

Claude 3.5 Haikuwon 1 · lost 8

O3 Miniwon 1 · lost 8

Claude 3.7 Sonnetwon 1 · lost 8

GLM-5.1won 1 · lost 8

Claude Sonnet 4.6won 1 · lost 8

Gemini 2.0 Flashwon 1 · lost 9

MiniMax M2.7won 1 · lost 9

GPT-5.1won 1 · lost 9

Gemini 3 Prowon 1 · lost 9

Claude Opus 5won 1 · lost 9

GPT-4o miniwon 1 · lost 10

Claude Haiku 4.5won 1 · lost 10

Claude Opus 4won 1 · lost 10

Qwen3.5 397B A17Bwon 1 · lost 10

Inklingwon 1 · lost 10

GPT-5.6 Lunawon 1 · lost 10

Grok 4.5won 1 · lost 10

GPT-5.6 Terrawon 1 · lost 10

Qwen3.8 Maxwon 1 · lost 10

Gemini 3.8 Flashwon 1 · lost 10

Claude Sonnet 4won 1 · lost 11

GPT-5won 1 · lost 11

Claude Opus 4.7won 1 · lost 11

Claude Opus 5.5won 1 · lost 11

GPT-5 nanowon 1 · lost 12

GPT-5 miniwon 1 · lost 12

GPT-4.1won 1 · lost 12

GPT-5.2won 1 · lost 12

Qwen Maxwon 0 · lost 8

Claude 3 Opuswon 0 · lost 8

GPT-4 Turbowon 0 · lost 8

O1 Miniwon 0 · lost 8

Qwen3 32Bwon 0 · lost 8

Claude 3.5 Sonnetwon 0 · lost 8

O1won 0 · lost 8

Hy3won 0 · lost 9

MiniMax M2won 0 · lost 10

Gemini 2.5 Flash-Litewon 0 · lost 10

Claude Sonnet 5won 0 · lost 10

MiMo V2.6 Prowon 0 · lost 10

Muse Spark 1.1won 0 · lost 10

O3won 0 · lost 11

Grok 4.7won 0 · lost 11

Grok 4.6won 0 · lost 11

Claude Opus 4.8won 0 · lost 11

GPT-4.1 miniwon 0 · lost 12

§ 3 · Sources

Where the numbers come from

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

LMArena Text ↗1 measureread 2026-11-10

LMArena Text 1303

BFCL ↗1 measureread 2026-11-10

BFCL v4 19.62%

Kagi LLM Benchmark ↗1 measureread 2026-11-10

Kagi LLM Benchmark 25.2%

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

Enkrypt · Jailbreak risk 10.9%Enkrypt · Harmful content risk 51.1%Enkrypt · CBRN risk 11.5%Enkrypt · Toxicity risk 5%Enkrypt · Bias risk 84%Enkrypt · Insecure code risk 44%

Vectara Hallucination Leaderboard ↗1 measureread 2026-11-10

Vectara · Factual consistency 93.6%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=gemma-3-4b-it)](https://publicai.io/model-index/m/gemma-3-4b-it)