The LLM benchmark aggregator.

Gemma 4 26B A4B

Google · 26B · A4B · open weights

Strongest in Factual grounding (#13 of 101), weakest in Jailbreak resistance (#223 of 272). Above par in 6 of 11 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5 and Claude Sonnet 5 and ahead of Kimi K2.6 and Qwen3.5 122B A10B.

§ 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.

Human preference60.7−6.8#80/3421/1

Agents48.1−20#140/2681/5

Knowledge work47.2−26.4#96/1781/2

Safety50.4−11.1#170/3372/3

Factual grounding62−8.6#13/1011/1

Secure code61.3−5.2#54/2741/1

Toxicity avoidance54.8−4.2#113/2721/1

Harm refusal47.3−14.3#204/3001/2

Fairness44−26#218/3001/2

Jailbreak resistance39.2−25.7#223/2721/1

§ 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 4 26B A4B, left for the other.

DeepSeek V3won 10 · lost 1

Grok 4.1 Fastwon 9 · lost 1

Command R Pluswon 8 · lost 1

Grok 3won 8 · lost 1

Smollm2 1.7B Instructwon 7 · lost 1

Gemma 7B Itwon 7 · lost 1

Nous Hermes 2 Mixtral 8x7b Dpowon 7 · lost 1

DeepSeek LLM 67B Chatwon 7 · lost 1

DBRX Instructwon 7 · lost 1

GPT-3.5 Turbowon 7 · lost 1

QwQ 32Bwon 7 · lost 1

Command R7Bwon 6 · lost 1

DeepSeek R1won 9 · lost 2

Llama 3.1 8B Instructwon 8 · lost 2

Command Awon 8 · lost 2

Qwen3 8Bwon 7 · lost 2

Llama 3.2 3B Instructwon 7 · lost 2

Qwen3 30B A3B Instructwon 7 · lost 2

Grok 4won 7 · lost 2

Gemma 4 E2Bwon 6 · lost 2

Gemma 4 E4Bwon 6 · lost 2

Gemma 2B Itwon 6 · lost 2

Zephyr 7B Betawon 6 · lost 2

Qwen1.5 14B Chatwon 6 · lost 2

Jamba 1.5 Miniwon 6 · lost 2

Magistral Mediumwon 6 · lost 2

GLM-4.5won 6 · lost 2

Mimo V2.6 Flashwon 6 · lost 2

GLM-4.6won 8 · lost 3

DeepSeek V3.2won 8 · lost 3

Gemini 2.5 Prowon 8 · lost 3

Jamba Mini 1.7won 5 · lost 2

Jamba Large 1.7won 5 · lost 2

Nova Microwon 7 · lost 3

Gemma 3 4B Itwon 7 · lost 3

GPT OSS 20Bwon 7 · lost 3

NVIDIA Nemotron 3 Super 120B A12Bwon 7 · lost 3

Phi-4 Mini Instructwon 6 · lost 3

Llama 3.2 1B Instructwon 6 · lost 3

Llama 4 Scout Instructwon 7 · lost 4

Gemma 4 31Bwon 7 · lost 4

Gemma 4 12Bwon 5 · lost 3

Ling 3.0 Flashwon 5 · lost 3

Nemotron 3 Ultra 550B A55Bwon 5 · lost 3

Grok Build 0.1won 5 · lost 3

Ling 3.0 Flash VLwon 5 · lost 3

Mistral 7B Instruct V0.2won 5 · lost 3

Granite 3.0 2B Instructwon 5 · lost 3

Granite 3.1 2B Instructwon 5 · lost 3

Gemma 2 2B Itwon 5 · lost 3

Llama 3 8B Instructwon 5 · lost 3

Gemma 2 9B Itwon 5 · lost 3

Mistral Small 24B Instructwon 5 · lost 3

Jamba 1.5 Largewon 5 · lost 3

Gemini 1.5 Prowon 5 · lost 3

Gemini 2.0 Flash Litewon 5 · lost 3

Qwen3 Next 80B A3B Thinkingwon 5 · lost 3

GPT-4won 6 · lost 4

GPT-4.1 Nanowon 6 · lost 4

Mistral Smallwon 6 · lost 4

O3 Miniwon 6 · lost 4

Gemini 2.5 Flashwon 6 · lost 4

Granite 3.2 8B Instructwon 4 · lost 3

Granite 3.1 8B Instructwon 5 · lost 4

Nova Litewon 5 · lost 4

Llama 4 Maverick Instructwon 6 · lost 5

Gemma 3 27B Itwon 6 · lost 5

Qwen3.5 122B A10Bwon 6 · lost 5

Kimi K2.6won 6 · lost 5

Phi 3 Mini 128k Instructwon 4 · lost 4

Phi 3 Mini 4k Instructwon 4 · lost 4

Phi 3 Small 8k Instructwon 4 · lost 4

Starling Lm 7B Betawon 4 · lost 4

Granite 3.0 8B Instructwon 4 · lost 4

Phi 3 Medium 4k Instructwon 4 · lost 4

Llama 3.1 Tulu 3 8Bwon 4 · lost 4

Claude 3 Haikuwon 4 · lost 4

Gemini 1.5 Flashwon 4 · lost 4

Gemini 2.0 Flashwon 4 · lost 4

Nova Prowon 5 · lost 5

Mistral Largewon 5 · lost 5

GPT-4o Miniwon 5 · lost 5

Claude 3.5 Haikuwon 5 · lost 5

GPT-4.1 Miniwon 5 · lost 5

O4 Miniwon 5 · lost 5

GPT-4.1won 5 · lost 5

Kimi K2won 5 · lost 5

Grok 4.3won 5 · lost 5

Gemini 3.6 Flashwon 5 · lost 5

Gemma 3 12B Itwon 5 · lost 6

GPT-5 Miniwon 5 · lost 6

C4ai Aya Expanse 8Bwon 4 · lost 5

GPT-4owon 4 · lost 5

Claude 3.7 Sonnetwon 4 · lost 5

Qwen3.5 27Bwon 4 · lost 5

Claude Opus 4won 4 · lost 5

Llama 3.3 Instruct Turbo (70B)won 3 · lost 4

Phi 4won 4 · lost 6

GPT-5 Nanowon 4 · lost 6

Qwen3 235B A22B Instructwon 4 · lost 6

GPT-5.2won 4 · lost 6

DeepSeek V4 Flashwon 4 · lost 6

MiniMax M3won 4 · lost 6

Mimo V2.5 Prowon 4 · lost 6

GLM-5.3 Flashwon 4 · lost 6

Qwen3.7 Maxwon 4 · lost 6

GLM-5.2won 4 · lost 6

Gemini 3.5 Flashwon 4 · lost 6

Qwen2 72B Instructwon 3 · lost 5

Llama 3 70B Instructwon 3 · lost 5

Claude 3 Sonnetwon 3 · lost 5

Qwen2.5 72B Instructwon 3 · lost 5

Qwen Maxwon 3 · lost 5

Claude 3 Opuswon 3 · lost 5

GPT-4 Turbowon 3 · lost 5

O1 Miniwon 3 · lost 5

MiniMax M2won 3 · lost 5

Claude 3.5 Sonnetwon 3 · lost 5

O1won 3 · lost 5

Qwen3 Vl 235B A22B Instructwon 3 · lost 5

GPT OSS 120Bwon 4 · lost 7

Claude Sonnet 4won 4 · lost 7

Claude Haiku 4.5won 4 · lost 7

C4ai Aya Expanse 32Bwon 3 · lost 6

O3won 3 · lost 6

Granite 4.0 H Smallwon 2 · lost 5

Gemini 2.5 Flash Litewon 2 · lost 5

Claude Opus 4.6won 2 · lost 5

Qwen3.5 35B A3Bwon 3 · lost 8

GPT-5won 3 · lost 8

Kimi K2.5won 3 · lost 8

GPT-6 Solwon 3 · lost 8

DeepSeek V4 Prowon 3 · lost 8

Kimi K2.7 Codewon 2 · lost 6

Gemma 2 27B Itwon 2 · lost 6

Claude Sonnet 4.6won 2 · lost 7

MiniMax M2.7won 2 · lost 8

GLM-5.1won 2 · lost 8

Muse Spark 1.1won 2 · lost 8

GPT-5.1won 2 · lost 9

Claude Sonnet 4.5won 2 · lost 9

GPT-5.5won 2 · lost 9

GPT-5.6 Solwon 2 · lost 9

Claude Opus 4.7won 2 · lost 9

Gemini 3 Prowon 1 · lost 6

GPT-6 Lunawon 1 · lost 9

Qwen3.5 397B A17Bwon 1 · lost 9

Grok 4.6won 1 · lost 9

GPT-5.6 Lunawon 1 · lost 9

Hy3won 1 · lost 9

Grok 4.5won 1 · lost 9

GPT-5.6 Terrawon 1 · lost 9

Qwen3.8 Maxwon 1 · lost 9

Mimo V2.6 Prowon 1 · lost 9

Claude Opus 4.8won 1 · lost 9

Gemini 3.8 Flashwon 1 · lost 9

Claude Opus 5.5won 1 · lost 9

GPT-5.4won 1 · lost 10

Grok 4.7won 0 · lost 10

Inklingwon 0 · lost 10

Claude Sonnet 5won 0 · lost 10

Claude Opus 5won 0 · lost 10

§ 3 · Sources

Where the numbers come from

4 publications, 9 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1438

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 553

Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-09-26

Enkrypt · Jailbreak risk 21.9%Enkrypt · Harmful content risk 5%Enkrypt · CBRN risk 37%Enkrypt · Toxicity risk 2.9%Enkrypt · Bias risk 87.1%Enkrypt · Insecure code risk 10.7%

Vectara Hallucination Leaderboard ↗1 measureread 2026-09-26

Vectara · Factual consistency 94.8%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=gemma-4-26b-a4b)](https://publicai.io/model-index/m/gemma-4-26b-a4b)