Which is better, GPT-4o or o3?

Across 54 shared benchmarks tracked on Vector Wire, GPT-4o leads on 5 and o3 leads on 48 — o3 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 o3

OpenAIvsOpenAI54 shared benchmarks5–48 head-to-head

Benchmark	GPT-4o	o3
AA Agentic Index	9.7	36.1
AA Intelligence	11.2	31.1
AA-LCR	55	73.3
AA-Omniscience	-10.5	-15.3
aider_polyglot	30.7	81.3
AIME 2024	9.3	91.6
AIME 2025	11.7	89.2
AIR-Bench 2024	62.4	84.5
anthropic_red_team	99.1	98.3
arc_agi_1	4.5	60.8
ARC-AGI-2	0	6.5
arena_vision	1162	1217
Artificial Analysis Coding Index	24.2	38.4
bbq	95.1	97.9
critpt	0	1.1
ERQA	35.2	64
Fortress	47.2	16
gdpval	0	12.8
GPQA Diamond	70.1	83.3
harmbench	82.9	98.4
HLE	5.3	20.3
IFBench	36	71.4
LiveCodeBench	38.3	84.7
livecodebench_easy	82.5	99.1
livecodebench_hard	4.5	66
livecodebench_medium	32.1	89.8
longbench_v2	48.1	58.8
math_500_em	74.6	98.1
MathVista	63.8	86.8
metr_hcast	25	64.7
metr_re_bench	0	15
metr_swaa	99.2	99.8
MMLU-Pro	74.7	85
MMMU	72.2	82.9
MMMU-Pro	59.9	76.4
multichallenge	40.3	60.4
MultiNRC	12.4	45.5
OmniScience Accuracy	23.7	38.6
OmniScience Non-Hallucination	62.1	12.9
PropensityBench	46.1	10.5
scicode	33.4	41
simple_safety_tests	98.5	99
simplebench	17.8	53.1
simpleqa	39.4	49.4
swe_bench_bash	21.6	58.4
SWE-bench Verified	38.8	69.1
TAU-bench (airline)	42.8	52
TAU-bench (retail)	60.3	73.9
Tau2 airline	45.5	64.8
Tau2 retail	63.4	80.2
Terminal-Bench Hard	8.3	37.1
VideoMMMU	61.2	83.3
xstest	97.3	97.3
τ²-Bench Telecom (AA run)	28.9	80.7

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.