The LLM benchmark aggregator.

GPT-5.4 Mini

OpenAI

Strongest in Factual grounding (#17 of 101), weakest in Core abilities (#196 of 204). Above par in 10 of 23 scopes. Among the models it meets almost everywhere, it finishes behind Muse Spark 1.1 and Claude Opus 5 and ahead of GPT-5 Mini and Grok 4 Fast.

§ 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 preference61.5−6#65/3421/1

Knowledge53.5−10#66/1381/2

Academic knowledge54.2−9.7#61/1231/1

Safety53.5−8#83/3371/3

Factual grounding61.3−9.3#17/1011/1

Agents51.1−17#105/2682/5

Knowledge work51.4−22.2#71/1782/2

Professional46.6−15#115/1681/1

Finance51.8−11.1#77/1511/1

Legal38.9−28.2#134/1511/1

Coding43.3−25.8#130/1652/5

Code generation38.1−31.9#64/771/2

Agentic coding44.7−24#107/1572/4

Reasoning43.5−23.4#137/1783/4

Science54.9−8.6#59/1221/1

Mathematics45−20.7#98/1402/2

Reasoning35.5−33.6#132/1402/3

Core abilities37.9−29.5#196/2042/3

Data analysis45.6−20.5#41/571/1

Instruction following35.8−37.9#50/571/1

Language34.4−37.1#53/571/1

General intelligence37−33.1#190/2042/3

§ 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-5.4 Mini, left for the other.

GPT-4o Miniwon 16 · lost 2

Mistral Medium 3.5won 14 · lost 2

GPT-4.1 Nanowon 15 · lost 3

Mistral Smallwon 11 · lost 3

Llama 4 Scout Instructwon 13 · lost 4

GPT-4owon 12 · lost 4

Llama 4 Maverick Instructwon 14 · lost 5

Mistral Large 3won 14 · lost 5

GPT OSS 20Bwon 11 · lost 4

Claude 3.5 Haikuwon 11 · lost 4

Command Awon 12 · lost 5

Gemini 2.0 Flashwon 11 · lost 5

Grok 4.3won 14 · lost 8

Qwen3.6 27Bwon 10 · lost 6

Gemini 3.1 Flash Litewon 11 · lost 7

GPT-5.4 Nanowon 14 · lost 9

MiniMax M2.7won 9 · lost 6

DeepSeek V3won 12 · lost 8

GPT OSS 120Bwon 11 · lost 8

Claude Haiku 4.5won 11 · lost 8

Mimo V2.5won 8 · lost 6

GPT-5 Nanowon 9 · lost 7

Grok 4.1 Fastwon 9 · lost 7

GLM-4.6won 10 · lost 8

Gemini 2.5 Flash Litewon 8 · lost 7

Grok 4 Fastwon 9 · lost 8

GPT-5 Miniwon 10 · lost 9

Grok Build 0.1won 7 · lost 7

Claude 3.5 Sonnetwon 7 · lost 7

Qwen3.5 Flashwon 7 · lost 7

DeepSeek R1won 9 · lost 10

O3 Miniwon 8 · lost 9

Claude 3.7 Sonnetwon 7 · lost 8

Grok 3won 7 · lost 8

GLM-4.7won 7 · lost 8

MiniMax M3won 10 · lost 12

Claude Sonnet 4won 9 · lost 11

Gemini 2.5 Prowon 9 · lost 11

GPT-4.1 Miniwon 8 · lost 10

GPT-4.1won 8 · lost 10

Kimi K2won 8 · lost 10

Nemotron 3 Ultra 550B A55Bwon 6 · lost 8

Gemma 4 31Bwon 7 · lost 10

MiniMax M2.5won 6 · lost 9

Qwen3 235B A22B Instructwon 7 · lost 11

Gemini 3.5 Flash Litewon 8 · lost 13

O4 Miniwon 7 · lost 12

Gemini 2.5 Flashwon 7 · lost 12

DeepSeek V3.2won 7 · lost 12

Kimi K2 Thinkingwon 5 · lost 9

O1won 5 · lost 9

Nemotron 3 Ultrawon 5 · lost 10

Grok 3 Miniwon 5 · lost 10

Claude Opus 4won 6 · lost 12

Kimi K2.7 Codewon 4 · lost 10

Inkling Smallwon 4 · lost 11

Qwen3.8 27Bwon 5 · lost 16

Qwen3.6 Pluswon 5 · lost 16

Grok 4won 4 · lost 13

Mimo V2.5 Prowon 4 · lost 13

Qwen3 Maxwon 3 · lost 11

GPT-6 Lunawon 4 · lost 15

Kimi K2.5won 4 · lost 15

Gemini 3 Flashwon 3 · lost 12

Inklingwon 4 · lost 18

GLM-5.3 Flashwon 4 · lost 18

Kimi K2.6won 4 · lost 19

Gemini 3.1 Prowon 4 · lost 19

GPT-5won 3 · lost 16

Claude Sonnet 4.5won 3 · lost 16

GPT-6 Solwon 3 · lost 17

Muse Sparkwon 2 · lost 12

Claude Opus 4.6won 3 · lost 18

GPT-5.2won 3 · lost 19

DeepSeek V4 Flashwon 3 · lost 19

Claude Opus 4.5won 3 · lost 19

Qwen3.7 Maxwon 3 · lost 19

GLM-5won 2 · lost 15

O3won 2 · lost 16

Gemini 3 Prowon 2 · lost 16

Grok 4.7won 2 · lost 17

DeepSeek V4.1 Flashwon 2 · lost 18

GPT-6 Astrawon 2 · lost 18

Gemini 3.5 Flashwon 2 · lost 20

Gemini 3.6 Flashwon 2 · lost 20

Gemini 3.8 Flashwon 2 · lost 20

DeepSeek V4 Prowon 2 · lost 21

Claude Sonnet 4.6won 2 · lost 21

GPT-5.6 Solwon 2 · lost 21

Mimo V2.6 Prowon 1 · lost 13

GLM-5.1won 1 · lost 15

GPT-5.1won 1 · lost 16

GLM-5.3won 1 · lost 20

Gemini 3.7 Flashwon 1 · lost 20

Grok 4.5won 1 · lost 21

GLM-5.2won 1 · lost 21

Qwen3.8 Maxwon 1 · lost 21

Claude Opus 4.8won 1 · lost 21

GPT-5.4won 1 · lost 22

GPT-5.5won 1 · lost 22

Claude Opus 4.7won 1 · lost 22

Muse Spark 1.3won 0 · lost 18

Claude Opus 5.5won 0 · lost 19

Muse Spark 1.2won 0 · lost 20

Kimi K3won 0 · lost 21

Claude Fable 5.1won 0 · lost 21

Claude Fable 5won 0 · lost 21

Grok 4.6won 0 · lost 22

GPT-5.6 Lunawon 0 · lost 22

Claude Sonnet 5won 0 · lost 22

GPT-5.6 Terrawon 0 · lost 22

Claude Opus 5won 0 · lost 22

Muse Spark 1.1won 0 · lost 22

§ 3 · Sources

Where the numbers come from

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

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1447

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 1000

AA-Briefcase ↗1 measureread 2026-09-26

AA-Briefcase 718

ARC-AGI-2 ↗1 measureread 2026-09-26

ARC-AGI-2 18.9%

LiveBench ↗8 measuresread 2026-09-26

LiveBench 66.4LiveBench · Reasoning 71.3LiveBench · Coding 71.6LiveBench · Agentic Coding 41.7LiveBench · Mathematics 78.5LiveBench · Data Analysis 70.8LiveBench · Language 71LiveBench · Instruction Following 59.8

Kagi LLM Benchmark ↗1 measureread 2026-09-26

Kagi LLM Benchmark 37.9%

Vals.ai ↗13 measuresread 2026-09-26

Vals · Legal Research Bench 12.5%Vals · CaseLaw 51.66%Vals · Harvey Legal Agent Benchmark 0%Vals · Finance Agent 45.36%Vals · CorpFin 60.92%Vals · TaxEval 71.22%Vals · MortgageTax 63.51%Vals · SWE-bench Verified 73% (Mini-SWE-agent)Vals · Vibe Code Bench 47.97% (OpenHands)Vals · Code Migration 12.94%Vals · GPQA Diamond 83.08%Vals · MMLU Pro 84.55%Vals · AIME 95.63%

Vectara Hallucination Leaderboard ↗1 measureread 2026-09-26

Vectara · Factual consistency 94.5%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=gpt-5-4-mini)](https://publicai.io/model-index/m/gpt-5-4-mini)