Request a pilot
‹ PublicAI Index

The LLM benchmark aggregator.

Gemma 4 12B

Google · 12B

Strongest in Secure code (#52 of 274), weakest in Fairness (#257 of 300). Above par in 6 of 16 scopes. Among the models it meets almost everywhere, it finishes behind Claude Sonnet 5 and Claude Opus 5.5 and ahead of DeepSeek V4 Flash and Llama 4 Maverick Instruct.

§ 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.

Agents46.5#158/268
Knowledge work44.8#108/178
Safety49#206/337
Secure code61.5#52/274
Toxicity avoidance54.6#122/272
Harm refusal48.4#182/300
Jailbreak resistance40.4#218/272
Fairness42.4#257/300
Core abilities47.8—/204✱
Long context41.7—/0✱
Reasoning52.6—/178✱
Mathematics53.9—/140✱
Expert reasoning56.8—/0✱
Coding48.8—/165✱
Agentic coding47.4—/157✱
Code generation51—/77✱

§ 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 12B, left for the other.

Command Awon 10 · lost 2
Gemma 4 E4Bwon 9 · lost 2
Granite 4.2 8Bwon 8 · lost 2
Grok 4.3won 11 · lost 3
Grok Build 0.1won 11 · lost 3
DeepSeek V3won 11 · lost 3
Mimo V2.6 Flashwon 7 · lost 3
Mistral Smallwon 7 · lost 3
GPT-4.1 Nanowon 9 · lost 4
GPT-4o Miniwon 9 · lost 5
Grok 4.1 Fastwon 7 · lost 4
Llama 4 Scout Instructwon 7 · lost 4
Gemini 2.0 Flashwon 7 · lost 4
MiniMax M3won 10 · lost 6
Claude 3.5 Haikuwon 7 · lost 5
Kimi K2.7 Codewon 8 · lost 6
Kimi K2.6won 8 · lost 6
Qwen3.7 Maxwon 8 · lost 6
DeepSeek R1won 8 · lost 6
GPT-4owon 6 · lost 5
Ling 3.0 Flashwon 6 · lost 5
Gemma 3 27B Itwon 6 · lost 5
GPT OSS 20Bwon 6 · lost 5
GLM-4.6won 7 · lost 6
GPT-4.1 Miniwon 7 · lost 6
DeepSeek V3.2won 7 · lost 6
DeepSeek V4 Flashwon 8 · lost 7
Qwen3.5 9Bwon 5 · lost 5
Gemini 2.5 Prowon 7 · lost 7
Gemma 4 31Bwon 7 · lost 8
Mimo V2.5 Prowon 6 · lost 7
Claude 3.5 Sonnetwon 5 · lost 6
Kimi K2 Thinkingwon 5 · lost 6
Inklingwon 7 · lost 9
GLM-5.2won 7 · lost 9
GLM-5.3 Flashwon 6 · lost 8
Gemini 3.6 Flashwon 6 · lost 8
Grok 4won 5 · lost 7
Qwen3.8 27Bwon 4 · lost 6
O3 Miniwon 5 · lost 8
Claude Haiku 4.5won 5 · lost 8
GPT OSS 120Bwon 5 · lost 8
Gemini 2.5 Flashwon 5 · lost 8
Kimi K2.5won 5 · lost 8
Gemini 3.5 Flashwon 5 · lost 9
Kimi K2won 4 · lost 8
GPT-4.1won 4 · lost 8
Claude Opus 4won 4 · lost 8
Claude Sonnet 4.5won 4 · lost 9
GPT-5 Miniwon 4 · lost 9
O1won 3 · lost 7
K2 Horizon 7Bwon 3 · lost 7
GPT-6 Lunawon 4 · lost 10
Muse Spark 1.1won 4 · lost 10
Gemini 3.8 Flashwon 4 · lost 10
Claude Sonnet 4won 4 · lost 10
Claude 3.7 Sonnetwon 3 · lost 8
DeepSeek V4 Prowon 4 · lost 11
GPT-5.1won 3 · lost 9
Claude Sonnet 4.6won 3 · lost 9
O4 Miniwon 3 · lost 10
Grok 4.5won 3 · lost 11
Grok 4.6won 3 · lost 11
Grok 4.7won 3 · lost 11
Qwen3.8 Maxwon 3 · lost 11
GPT-5.6 Lunawon 3 · lost 13
MiniMax M2.7won 2 · lost 10
GLM-5.1won 2 · lost 10
Qwen3 235B A22B Instructwon 2 · lost 10
GPT-5.2won 2 · lost 11
O3won 2 · lost 11
GPT-6 Solwon 2 · lost 12
GPT-5.6 Solwon 2 · lost 13
Claude Opus 4.6won 1 · lost 9
GPT-5 Nanowon 1 · lost 9
Mimo V2.6 Prowon 1 · lost 12
Claude Opus 4.8won 1 · lost 13
Claude Opus 4.7won 1 · lost 13
GPT-5.4won 1 · lost 13
GPT-5.5won 1 · lost 13
Claude Opus 5won 1 · lost 14
GPT-5.6 Terrawon 1 · lost 15
GPT-5won 0 · lost 13
Claude Opus 5.5won 0 · lost 14
Claude Sonnet 5won 0 · lost 16

§ 3 · Sources

Where the numbers come from

3 publications, 13 figures. Every one links to the page it was read from.

GDPval-AA ↗1 measureread 2026-09-26
GDPval-AA 413
Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-09-26
Enkrypt · Jailbreak risk 20.9%Enkrypt · Harmful content risk 4.4%Enkrypt · CBRN risk 34.5%Enkrypt · Toxicity risk 3.1%Enkrypt · Bias risk 89.7%Enkrypt · Insecure code risk 10.2%
Terminal-Bench 2.1 27.3%SciCode 38.2%AA-LCR 61.7%SWE-bench Verified 30.6%HMMT Feb 2026 63.1%HLE 15.7%

Badge

PublicAI Index badge for Gemma 4 12B[![PublicAI Index](https://publicai.io/model-index/badge?model=gemma-4-12b)](https://publicai.io/model-index/m/gemma-4-12b)

✱ Placed by a one-off publication, not a board that re-ran the model. ✱✱ marks a figure the model’s own publisher printed, which counts for less again. Scores are 0–100 on the PublicAI Index scale, 50 = the average of the models each source lists. Scores belong to their publishers.

Snapshot 2026-09-26 · the full index · JSON API