Which is better, GPT-4o or GPT-4o mini?

Across 59 shared benchmarks tracked on Vector Wire, GPT-4o leads on 55 and GPT-4o mini leads on 2 — GPT-4o leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

GPT-4o vs GPT-4o mini

OpenAIvsOpenAI59 shared benchmarks55–2 head-to-head

Benchmark	GPT-4o	GPT-4o mini
AA Agentic Index	9.7	1
AA Intelligence	11.2	7
AI2D	94.2	75.2
AI2D (Acc)	84.6	77.8
aider_polyglot	30.7	3.6
AIR-Bench 2024	62.4	56.3
anthropic_red_team	99.1	98.3
ARC-AGI-2	0	0
Artificial Analysis Coding Index	24.2	11.4
bbq	95.1	88.2
BLINK (Acc)	68	53.6
DROP	83.4	79.7
Fortress	47.2	48.1
gaia_level1	39.8	7.5
gaia_level2	27	4.4
gaia_level3	14.3	0
gaia_overall	28.9	4.7
gdpval	0	0
GPQA Diamond	70.1	42.6
GSM8K	95.6	84.3
harmbench	82.9	84.9
HLE	5.3	4.2
humaneval	90.6	87.2
IFBench	36	31
InfoVQA (Acc)	80.7	57.9
LiveCodeBench	38.3	35.5
livecodebench_easy	82.5	81.7
livecodebench_hard	4.5	4.3
livecodebench_medium	32.1	25.2
LongVideoBench (val)	66.7	58.2
MATH	85.3	80.2
MathVista	63.8	56.7
mgsm	90.5	87
MLVU-MCQ (Acc)	64.6	48.1
MMBench-EN-v1.1 (Acc)	83.1	77.1
MMLongBench-DOC (Acc)	42.8	29
mmlu	88.1	81.8
MMMU	72.2	59.4
MMMU (val) (Pass@1)	69.1	52.1
MMMU-Pro	59.9	41.5
MMStar (Acc)	64.7	54.8
MMVet (Pass@1)	69.1	66.9
MMVU-Val (Pass@1)	67.4	61.6
narrativeqa	80.4	76.8
naturalquestions_closedbook	50.1	38.5
OCRBench (Acc)	815	785
OpenBookQA	96.8	92
RealWorldQA (Acc)	75.4	67.1
scicode	33.4	22.9
ScreenSpot-V2	18.1	6.9
simple_safety_tests	98.5	97.8
simplebench	17.8	10.7
simpleqa	39.4	9.9
SWE-bench Verified	38.8	8.7
TOMATO	37.7	28.8
Video-MME	77.2	68.9
Video-MME (w/o sub.)	71.9	64.8
WindowsAgentArena (Pass@1)	9.4	2.7
xstest	97.3	96

Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.