The LLM benchmark aggregator.

GPT-5.4 nano

OpenAI

Strongest in Factual grounding (#2 of 104), weakest in Core abilities (#180 of 199). Above par in 7 of 23 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 4.8 and Claude Fable 5 and ahead of O3 Mini and Gemini 2.0 Flash.

§ 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.

Safety55.4−6#41/3291/3

Factual grounding67.4−3.3#2/1041/1

Reasoning51.4−14#90/1953/5

Mathematics55.3−8.5#49/1522/2

Science51−12.5#74/1221/1

Reasoning43−26#97/1471/3

Human preference56.8−12#132/3541/1

Coding43.8−22.4#138/1792/6

Agentic coding43.5−23.4#133/1732/5

Professional45.5−16.9#149/1941/2

Medical48.4−18.1#97/1491/1

Finance47.5−14.9#111/1621/1

Legal39.7−26.2#150/1731/1

Agents48.8−17.6#154/2982/6

Knowledge work48.1−22.6#113/2082/2

Knowledge40.1−22.4#178/1972/3

Document parsing39−24.3#70/831/1

Academic knowledge44.9−19#98/1231/1

Core abilities41.4−23.9#180/1992/4

Instruction following48.6−25.2#34/621/1

Language29.5−37.1#60/621/1

General intelligence43.6−25.8#131/1781/3

§ 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 nano, left for the other.

NVIDIA Nemotron 3.5 Lightning 30B A3Bwon 16 · lost 0

GPT-4.1 nanowon 17 · lost 2

GPT-4o miniwon 16 · lost 2

Llama 4 Scout Instructwon 15 · lost 3

Llama 3.3 70B Instructwon 12 · lost 3

Mistral Medium 3.1won 11 · lost 3

Mistral Medium 3.5won 13 · lost 4

GPT OSS 20Bwon 12 · lost 4

GPT-4owon 13 · lost 5

Command Awon 13 · lost 5

GPT-5 nanowon 12 · lost 5

Llama 4 Maverick Instructwon 14 · lost 6

Claude 3.5 Haikuwon 10 · lost 5

Gemma 4 31Bwon 11 · lost 6

Claude Haiku 4.5won 11 · lost 8

Gemini 2.5 Flash-Litewon 9 · lost 7

Gemini 2.0 Flashwon 9 · lost 8

O3 Miniwon 9 · lost 8

GPT OSS 120Bwon 10 · lost 10

DeepSeek V3won 10 · lost 10

Mistral Large 3won 10 · lost 10

O4 Miniwon 9 · lost 10

MiniMax M2.1won 7 · lost 9

Claude 3.7 Sonnetwon 7 · lost 9

GPT-4.1 miniwon 8 · lost 11

Gemini 3.1 Flash-Litewon 8 · lost 12

GPT-5.4 miniwon 8 · lost 13

Kimi K2won 7 · lost 12

DeepSeek R1won 7 · lost 12

Grok 4.3won 7 · lost 12

Claude 3.5 Sonnetwon 5 · lost 9

Grok 3 Miniwon 6 · lost 11

DeepSeek V3.2won 7 · lost 13

Qwen3.5 Flashwon 5 · lost 10

Grok 3won 6 · lost 12

GPT-5 miniwon 7 · lost 14

NVIDIA Nemotron 3 Ultra 550B A55Bwon 7 · lost 14

Gemini 2.5 Flashwon 6 · lost 13

Qwen3 235B A22B Instructwon 6 · lost 13

GLM-4.6won 6 · lost 13

Grok 4.20won 5 · lost 11

Claude Sonnet 4won 6 · lost 14

Kimi K2.6won 6 · lost 14

MiniMax M2.7won 5 · lost 12

Grok 4.1 Fastwon 5 · lost 12

Claude Opus 4.1won 4 · lost 10

Inklingwon 6 · lost 15

MiniMax M2.5won 5 · lost 13

Grok 4 Fastwon 5 · lost 13

Mistral Large 4won 5 · lost 13

Qwen3.6 Pluswon 5 · lost 13

Qwen3 Maxwon 4 · lost 11

MiMo V2.5won 4 · lost 11

GPT-4.1won 5 · lost 14

MiMo V2.5 Prowon 4 · lost 12

MiMo V2.6 Prowon 4 · lost 12

Gemini 2.5 Prowon 5 · lost 15

Gemini 3 Flashwon 4 · lost 13

Gemini 3.1 Prowon 5 · lost 17

Claude Opus 4won 4 · lost 14

GLM-4.7won 4 · lost 14

Kimi K2 Thinkingwon 3 · lost 12

O1won 3 · lost 12

MiMo V2.6 Flashwon 3 · lost 12

GPT-6 Lunawon 4 · lost 16

Kimi K2.5won 4 · lost 16

Gemini 3.5 Flash-Litewon 4 · lost 16

Qwen3.8 27Bwon 4 · lost 17

MiniMax M3won 4 · lost 17

Claude Sonnet 4.6won 4 · lost 17

Claude Opus 4.6won 4 · lost 17

GLM-5.3 Flashwon 4 · lost 18

Inkling Smallwon 3 · lost 15

Qwen3.7 Maxwon 3 · lost 17

GLM-4.5won 2 · lost 12

Step 5won 2 · lost 12

DeepSeek V4 Flashwon 3 · lost 18

Claude Opus 4.5won 3 · lost 18

DeepSeek V4.1 Flashwon 3 · lost 18

GLM-5.2won 3 · lost 18

GPT-5.6 Lunawon 3 · lost 19

DeepSeek V4 Prowon 3 · lost 19

GLM-5.1won 2 · lost 15

Gemini 4 Argonwon 2 · lost 15

Grok 4won 2 · lost 16

Grok 4.7won 2 · lost 16

GLM-5won 2 · lost 16

Gemini 3 Prowon 2 · lost 17

GPT-6 Solwon 2 · lost 18

GPT-5.2won 2 · lost 19

Grok 4.5won 2 · lost 19

GPT-6 Astrawon 2 · lost 19

Gemini 3.5 Flashwon 2 · lost 19

Gemini 3.6 Flashwon 2 · lost 19

Qwen3.8 Maxwon 2 · lost 19

Claude Opus 5.5won 2 · lost 19

Gemini 3.8 Flashwon 2 · lost 20

GPT-5.4won 2 · lost 21

GPT-5.5won 2 · lost 21

GPT-5.6 Solwon 2 · lost 21

Claude Opus 4.7won 2 · lost 21

O3won 1 · lost 17

GPT-5.1won 1 · lost 17

Muse Spark 1.2won 1 · lost 17

Claude Haiku 5.5won 1 · lost 18

Muse Spark 1.1won 1 · lost 18

GPT-5won 1 · lost 19

Claude Sonnet 4.5won 1 · lost 19

GLM-5.3won 1 · lost 19

Kimi K3won 1 · lost 19

Gemini 3.7 Flashwon 1 · lost 19

Claude Opus 5won 1 · lost 19

Grok 4.6won 1 · lost 20

Claude Sonnet 5won 1 · lost 20

GPT-5.6 Terrawon 1 · lost 21

Muse Sparkwon 0 · lost 15

Muse Spark 1.3won 0 · lost 16

Muse Wonderwon 0 · lost 16

Claude Sonnet 5.5won 0 · lost 19

GPT-6.1 Solwon 0 · lost 19

Claude Fable 5.1won 0 · lost 21

Claude Fable 5won 0 · lost 21

Claude Opus 4.8won 0 · lost 22

§ 3 · Sources

Where the numbers come from

9 publications, 35 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-11-10

LMArena Text 1401

GDPval-AA ↗1 measureread 2026-11-10

GDPval-AA 955

AA-Briefcase ↗1 measureread 2026-11-10

AA-Briefcase 674

ARC-AGI-2 ↗1 measureread 2026-11-10

ARC-AGI-2 5.7%

LiveBench 1.0 ↗8 measuresread 2026-11-10

LiveBench 1.0 69.6LiveBench 1.0 · Reasoning 81.1LiveBench 1.0 · Coding 70.8LiveBench 1.0 · Agentic Coding 46.8LiveBench 1.0 · Mathematics 91LiveBench 1.0 · Data Analysis 67.6LiveBench 1.0 · Language 62.5LiveBench 1.0 · Instruction Following 67.2

Kagi LLM Benchmark ↗1 measureread 2026-11-10

Kagi LLM Benchmark 39.7%

Vals.ai ↗15 measuresread 2026-11-10

Vals · Legal Research Bench 6.25%Vals · CaseLaw 51.88%Vals · LegalBench 77.92%Vals · Finance Agent 38.22%Vals · CorpFin 61.19%Vals · TaxEval 67.42%Vals · MortgageTax 59.1%Vals · MedCode 41.03%Vals · MedScribe 77.09%Vals · SWE-bench Verified 69.8% (Mini-SWE-agent)Vals · Vibe Code Bench 26.1% (OpenHands)Vals · Code Migration 14.47%Vals · GPQA Diamond 77.53%Vals · MMLU Pro 77.17%Vals · AIME 88.75%

ParseBench ↗6 measuresread 2026-11-10

ParseBench · Overall 42.24%ParseBench · Tables 60.28%ParseBench · Charts 16.67%ParseBench · Content faithfulness 79.19%ParseBench · Semantic formatting 48.71%ParseBench · Visual grounding 6.37%

Vectara Hallucination Leaderboard ↗1 measureread 2026-11-10

Vectara · Factual consistency 96.9%

Badge

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