Which is better, GPT-4o or GPT-5?

Across 52 shared benchmarks tracked on Vector Wire, GPT-4o leads on 6 and GPT-5 leads on 46 — GPT-5 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-5

OpenAIvsOpenAI52 shared benchmarks6–46 head-to-head

Benchmark	GPT-4o	GPT-5
AA Agentic Index	9.7	49.7
AA Intelligence	11.2	35.3
AA-LCR	55	76.3
AA-Omniscience	-10.5	-8.7
aider_polyglot	30.7	88
AIME 2025	11.7	95
AIR-Bench 2024	62.4	87.7
arc_agi_1	4.5	56.2
ARC-AGI-2	0	7.5
arena_vision	1162	1212
Artificial Analysis Coding Index	24.2	38.9
critpt	0	5.7
ERQA	35.2	65.7
Fortress	47.2	17
gaia_level1	39.8	78.5
gaia_level2	27	64.8
gaia_level3	14.3	46.9
gaia_overall	28.9	66.1
gdpval	0	32.5
GPQA Diamond	70.1	87.3
GSM8K	95.6	-1.8
harmbench	82.9	70
HLE	5.3	28.5
humaneval	90.6	93.4
IFBench	36	73.1
LiveCodeBench	38.3	85
LiveCodeBench v6	30.9	87
metr_hcast	25	68.4
metr_re_bench	0	40
metr_swaa	99.2	100
mmlu_redux	88	95.3
MMLU-Pro	74.7	87.5
MMMU	72.2	84.2
MMMU-Pro	59.9	78.4
multichallenge	40.3	71.1
MultiNRC	12.4	52.1
OmniScience Accuracy	23.7	40.3
OmniScience Non-Hallucination	62.1	21
scicode	33.4	43
SEAL VISTA	34.9	49.7
simplebench	17.8	56.7
swe_bench_bash	21.6	65
SWE-bench Verified	38.8	74.9
Tau2 airline	45.5	62.6
Tau2 retail	63.4	81.1
Terminal-Bench Hard	8.3	37.9
vectara_answer_rate	93.8	99.9
vectara_avg_summary_length	86.6	162.7
vectara_factual_consistency	90.4	85.3
vectara_hallucination_rate ↓	9.6	14.7
VideoMMMU	61.2	84.6
τ²-Bench Telecom (AA run)	28.9	86.5

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.