Request a pilot
‹ PublicAI Index

The LLM benchmark aggregator.

Gemma 4 31B

Google · 31B · open weights

Strongest in Factual grounding (#34 of 101), weakest in Fairness (#270 of 300). Above par in 12 of 28 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5 and Claude Opus 5.5 and ahead of MiniMax M3 and GPT-4.1 Mini.

§ 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 abilities53.7#56/204
General intelligence55.3#58/204
Long context49.8—/0✱
Human preference62.1#60/342
Human preference62.1#60/342
Professional47.8#106/168
Finance50.6#85/151
Legal43.4#115/151
Safety49.7#188/337
Factual grounding56.5#34/101
Secure code60.6#62/274
Toxicity avoidance52.7#160/272
Harm refusal48.5#179/300
Jailbreak resistance42.2#212/272
Fairness41.2#270/300
Agents44.6#188/268
Knowledge work43#116/178
Tool use45.8—/81✱
Reasoning49.4—/178✱
Science49.7—/122✱
Expert reasoning46—/0✱
Coding50.8—/165✱
Agentic coding50—/157✱
Code generation54.2—/77✱
Knowledge44.1—/138✱
Factuality41.2—/0✱
Decisions60.2—/85✱
Routing & classification63.7—/85✱

§ 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 31B, left for the other.

Command Awon 16 · lost 5
GPT-4o Miniwon 15 · lost 5
Llama 4 Scout Instructwon 15 · lost 5
DeepSeek V3won 16 · lost 6
Mistral Smallwon 13 · lost 5
GPT-4.1 Nanowon 15 · lost 6
Claude 3.5 Haikuwon 12 · lost 6
Gemini 2.0 Flashwon 12 · lost 6
Gemini 3.1 Flash Litewon 12 · lost 6
Grok 4.3won 14 · lost 7
Grok 3won 11 · lost 6
GLM-4.6won 14 · lost 8
GPT OSS 20Bwon 11 · lost 7
GPT-5.4 Nanowon 10 · lost 7
GPT-5.4 Miniwon 10 · lost 7
Grok 4.1 Fastwon 11 · lost 8
DeepSeek R1won 12 · lost 9
Claude Haiku 4.5won 12 · lost 10
DeepSeek V3.2won 12 · lost 10
Kimi K2.6won 12 · lost 10
GPT-5 Nanowon 10 · lost 9
O3 Miniwon 10 · lost 9
GPT-4.1 Miniwon 11 · lost 10
MiniMax M3won 13 · lost 12
GPT-4owon 9 · lost 9
Mimo V2.5 Prowon 10 · lost 10
Gemini 2.5 Flashwon 11 · lost 11
DeepSeek V4 Flashwon 11 · lost 11
GPT OSS 120Bwon 10 · lost 11
Kimi K2won 10 · lost 11
Nemotron 3 Ultrawon 8 · lost 9
Gemini 3.6 Flashwon 9 · lost 12
Gemini 2.5 Prowon 9 · lost 13
Grok 4won 8 · lost 12
Qwen3.7 Maxwon 8 · lost 13
Claude Sonnet 4won 8 · lost 14
Gemini 3.1 Prowon 6 · lost 11
Claude Opus 4won 7 · lost 13
MiniMax M2.7won 6 · lost 12
GPT-4.1won 7 · lost 14
Kimi K2.5won 7 · lost 14
GLM-5.3 Flashwon 7 · lost 14
Gemini 3.5 Flashwon 7 · lost 14
Inklingwon 8 · lost 17
O4 Miniwon 7 · lost 15
Qwen3.8 27Bwon 6 · lost 13
Claude 3.7 Sonnetwon 5 · lost 12
Claude Opus 4.5won 5 · lost 12
Qwen3 235B A22B Instructwon 6 · lost 15
GPT-6 Lunawon 6 · lost 15
GPT-5 Miniwon 6 · lost 16
GPT-5.2won 6 · lost 16
Claude Sonnet 4.5won 6 · lost 16
Claude Sonnet 4.6won 5 · lost 15
GLM-5.2won 6 · lost 19
Claude Opus 4.6won 4 · lost 14
DeepSeek V4 Prowon 5 · lost 18
Muse Spark 1.1won 4 · lost 17
Gemini 3 Prowon 3 · lost 15
Grok 4.7won 3 · lost 16
GPT-5.1won 3 · lost 16
O3won 3 · lost 18
GPT-5won 3 · lost 18
Grok 4.5won 3 · lost 18
GLM-5.1won 2 · lost 16
GPT-5.6 Lunawon 3 · lost 24
GPT-6 Solwon 2 · lost 18
Qwen3.8 Maxwon 2 · lost 19
Gemini 3.8 Flashwon 2 · lost 19
GPT-5.5won 2 · lost 20
Claude Opus 4.7won 2 · lost 20
GPT-5.6 Solwon 2 · lost 21
GPT-5.6 Terrawon 2 · lost 23
DeepSeek V4.1 Flashwon 1 · lost 17
Mimo V2.6 Prowon 1 · lost 17
Grok 4.6won 1 · lost 20
Claude Opus 4.8won 1 · lost 20
GPT-5.4won 1 · lost 21
Claude Sonnet 5won 1 · lost 24
Claude Opus 5.5won 0 · lost 19
Claude Opus 5won 0 · lost 22

§ 3 · Sources

Where the numbers come from

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

LMArena Text ↗1 measureread 2026-09-26
LMArena Text 1453
GDPval-AA ↗1 measureread 2026-09-26
GDPval-AA 606
AA-Briefcase ↗1 measureread 2026-09-26
AA-Briefcase 367
Kagi LLM Benchmark ↗1 measureread 2026-09-26
Kagi LLM Benchmark 63.5%
Vals.ai ↗2 measuresread 2026-09-26
Vals · CaseLaw 52.63%Vals · MortgageTax 61.37%
Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-09-26
Enkrypt · Jailbreak risk 19.4%Enkrypt · Harmful content risk 6.1%Enkrypt · CBRN risk 33.3%Enkrypt · Toxicity risk 4.4%Enkrypt · Bias risk 91.5%Enkrypt · Insecure code risk 12%
Vectara · Factual consistency 92.6%
tau3-Banking 14.8%Terminal-Bench 2.1 43.4%SciCode 43.4%Humanity's Last Exam (without tools) 23.6%GPQA Diamond 85.7%CritPt 1.4%AA-LCR 68.3%AA-Omniscience Accuracy 20%AA-Omniscience Non-Hallucination 15%
Decision correct 77%

Badge

PublicAI Index badge for Gemma 4 31B[![PublicAI Index](https://publicai.io/model-index/badge?model=gemma-4-31b)](https://publicai.io/model-index/m/gemma-4-31b)

✱ 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