GPT-4.1 nano

GPT-4.1 nano.

OpenAI Released April 14th, 2025 Compare with others

Benchmarks with a score

Ranked first, out of five or more models

In the top three

OpenAI Apr 2025

GPT-4.1 nano

Closed weights

126.8 #168 of 256

Epoch Capabilities Index

Self-Reported Index 115.5 #365 of 404

GPQA Diamond 48.9% #160 /229
FrontierMath 1.03% #52 /55
ProfBench 33.2% #70 /72

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
Vectara Hallucination Leaderboard (HHEM)	2.00%	#25 of 148	Official leaderboard
MMLU	80.1%	#77 of 131	Lab model card
LMArena (Elo)	1,322	#207 of 335	Official leaderboard
Wer wird Millionär? LLM Benchmark	389.0	#51 of 77	Official leaderboard
Internal API instruction following (OpenAI)	31.6%	#6 of 8	Official leaderboard
EQ-Bench 3	1,050	#61 of 78	Official leaderboard
MMBench-CN v1.1	60.4%	#140 of 178	Official leaderboard
HELM Arabic	70.2	#24 of 28	Official leaderboard
MMMLU	66.9%	#61 of 71	Lab model card
Creative Writing v3	50.9	#105 of 120	Official leaderboard
IFEval	74.5%	#120 of 135	Lab model card
Multi-IF	57.2%	#25 of 28	Lab model card
OpenAI MRCR	36.6%	#9 of 10	Lab model card
RewardBench 2	48.5%	#38 of 42	Official leaderboard
Arena-Hard v2.0	13.7%	#21 of 23	Official leaderboard
Longform Creative Writing (EQ-Bench)	32.4	#109 of 119	Official leaderboard
Phare	57.5%	#58 of 62	Official leaderboard
GraphWalks	25%	#26 of 27	Lab model card
Fiction.LiveBench	25%	#56 of 57	Fiction.live leaderboard, via Epoch AI
SimpleQA Verified	6.00%	#94 of 94	Epoch AI run
MultiChallenge	15%	#42 of 42	Lab model card
SimpleQA	7.60%	#16 of 16	Official leaderboard
COLLIE	42.5%	#12 of 12	Lab model card

Math and science

Benchmark	Score	Rank	Source
MATH Level 5	70%	#34 of 94	Epoch AI run
GPQA Main	50.7%	#36 of 60	Official leaderboard
OTIS Mock AIME 2024-2025	28.9%	#148 of 212	Epoch AI run
GPQA Diamond	48.9%	#160 of 229	Epoch AI run
MATH-500	80.2%	#31 of 38	Official leaderboard
DTBench	52.5%	#158 of 180	Official leaderboard
Misguided Attention	26.6%	#40 of 45	Official leaderboard
AIME 2024	29.4%	#80 of 88	Lab model card
FrontierMath	1.03%	#52 of 55	Official leaderboard
ARC-AGI-2	0%	#96 of 98	Official leaderboard
ARC-AGI	0%	#95 of 95	Official leaderboard

Coding

Benchmark	Score	Rank	Source
BigCodeBench-Hard	30.1%	#21 of 96	Official leaderboard
DesignArena Website	966	#137 of 151	Official leaderboard
WeirdML	19%	#103 of 112	Official leaderboard
DesignArena Game Dev	961	#132 of 143	Official leaderboard
DesignArena Code	953	#135 of 143	Official leaderboard
DesignArena 3D Design	908	#129 of 136	Official leaderboard
DesignArena UI Component	907	#132 of 139	Official leaderboard
DesignArena Data Visualization	889	#117 of 118	Official leaderboard
Aider Polyglot	9.80%	#30 of 30	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
BFCL V4	33%	#35 of 65	Official leaderboard
Galileo Agent Leaderboard v2	38%	#13 of 21	Official leaderboard
Berkeley Function Calling Leaderboard (BFCL)	53%	#42 of 67	Official leaderboard
Galileo Agent Leaderboard	0.80	#22 of 29	Official leaderboard
SnakeBench	16	#166 of 188	Official leaderboard
LisanBench	54.0	#121 of 133	Official leaderboard
DuelLab GameBench 2	11.1	#54 of 57	Official leaderboard
Chess Puzzles	0%	#93 of 93	Epoch AI run
MCPMark	0%	#28 of 28	Official leaderboard
tau-bench	18.3%	#23 of 23	Lab model card
ComplexFuncBench	5.70%	#3 of 3	Lab model card

Professional work

Benchmark	Score	Rank	Source
LEXam	43.7%	#16 of 30	Official leaderboard
MortgageTax	52.8%	#74 of 87	Official leaderboard
TaxEval v2	60.8%	#113 of 129	Official leaderboard
CorpFin v2	42.1%	#109 of 119	Official leaderboard
LegalBench	61.1%	#130 of 138	Official leaderboard
ProfBench	33.2%	#70 of 72	Official leaderboard
Vals Finance Agent v1	2.47%	#46 of 47	Official leaderboard

Vision, audio and robotics

Benchmark	Score	Rank	Source
MM-Vet	62.8%	#45 of 164	Official leaderboard
CharXiv (DQ)	73.9%	#19 of 67	Official leaderboard
SEED-Bench-2-Plus	67%	#40 of 136	Official leaderboard
A-Bench	72.9%	#52 of 116	Official leaderboard
MMStar	47.5%	#115 of 173	Official leaderboard
CCBench (MMBench Chinese Culture)	31.2%	#106 of 152	Official leaderboard
MMBench	68.2%	#120 of 155	Official leaderboard
MMBench-EN v1.1	64.4%	#147 of 178	Official leaderboard
Math-VR	9.10%	#22 of 26	Official leaderboard
VHELM	29.1	#35 of 41	Official leaderboard
Vision Arena	1,089	#117 of 136	Official leaderboard
MMMU	55.4%	#90 of 103	Lab model card
MathVista	56.2%	#62 of 70	Lab model card
VISTA	26.6%	#48 of 54	Official leaderboard
vCX Benchmark	64.4%	#40 of 42	Official leaderboard
CharXiv	40.5%	#75 of 75	Lab model card
MADQA	19.5%	#18 of 18	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
Anthropic Red Team	99.6%	#14 of 75	Official leaderboard
HELM Safety	93.8	#28 of 75	Official leaderboard
SimpleSafetyTests	99%	#36 of 75	Official leaderboard
AIR-Bench 2024	61.5%	#51 of 82	Official leaderboard
Human Pathogen Capabilities Test	35.2%	#76 of 86	Official leaderboard
nocot-bench (NCRI)	50.2	#165 of 185	Official leaderboard
World-Class Bio	17.2%	#70 of 74	Official leaderboard
Molecular Biology Capabilities Test	24.5%	#88 of 90	Official leaderboard

Other

Benchmark	Score	Rank	Source
OpenVLM Leaderboard	57.5	#96 of 173	Official leaderboard
HELM Capabilities	61.6	#39 of 60	Official leaderboard
Conceptual Reasoning Index	12.3	#144 of 151	Official leaderboard
ACCoRD	24.3	#145 of 152	Official leaderboard
LMCA	5.46%	#141 of 147	Official leaderboard

Its lab's numbers put it at 115.5, 11.3 points below its ECI

The Self-Reported Index fits one score to the 15 capability results OpenAI reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 112.8 to 118.3. How we compute it.

Its lab's numbers next to 5 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
IFEval	74.5% IFEval	84.3% Leaderboard, strict, share of instructions followed	−9.8 pts
GPQA Diamond	50.3% GPQA Diamond (model-extracted answers)	48.9% Epoch AI run	+1.4 pts
Aider Polyglot	9.80% Aider's polyglot: whole 9.8 (diff 6.2)	8.90% Leaderboard, aider, whole edits	+0.9 pts
Internal API instruction following (OpenAI)	31.6% Internal API instruction following (hard)	31.6% Leaderboard	Same
CharXiv	40.5% CharXiv-R 40.5 (CharXiv-D 73.9)	40.5% Leaderboard	Same

OpenAI's models over time

OpenAI's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Introducing GPT-4.1 in the API (OpenAI) Apr 14, 2025

AIME '24 29.4
GPQA Diamond (model-extracted answers) 50.3
MMLU 80.1
Multilingual MMLU 66.9
SWE-Lancer (IC-Diamond subset) $9K (3.7%) 3.7
Aider's polyglot: whole 9.8 (diff 6.2) 9.8
Internal API instruction following (hard) 31.6
MultiChallenge (GPT-4o grader) 15.0 (o3-mini grader 31.1) 15
COLLIE 42.5
IFEval 74.5
Multi-IF 57.2
OpenAI-MRCR: 2 needle 128k 36.6 (1M: 12.0) 36.6
Graphwalks bfs <128k 25.0 (parents <128k 9.4) 25
MMMU 55.4
MathVista 56.2
CharXiv-R 40.5 (CharXiv-D 73.9) 40.5
ComplexFuncBench 5.7
Mean of Taubench airline 14.0 / retail 22.6 18.3