Which is better, Claude Opus 4.5 or GPT-4o?

Across 51 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 46 and GPT-4o leads on 5 — Claude Opus 4.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Opus 4.5 vs GPT-4o

AnthropicvsOpenAI51 shared benchmarks46–5 head-to-head

Benchmark	Claude Opus 4.5	GPT-4o
AA Agentic Index	59.6	9.7
AA Intelligence	41.9	11.2
AA-LCR	76	55
AA-Omniscience	14	-10.5
AIME 2025	92.8	11.7
arc_agi_1	82	4.5
ARC-AGI-2	37.6	0
Artificial Analysis Coding Index	47.8	24.2
C-Eval	92.2	76
critpt	4.6	0
ERQA	46.8	35.2
Fortress	13.6	47.2
gdpval	47.3	0
GPQA Diamond	87	70.1
HLE	30.8	5.3
IFBench	58	36
LiveCodeBench	87	38.3
LiveCodeBench v6	84.8	30.9
longbench_v2	64.4	48.1
mathvision	77.1	30.4
MathVista	80.2	63.8
metr_hcast	73.4	25
metr_re_bench	40	0
metr_swaa	99.5	99.2
mmlu_redux	95.6	88
MMLU-Pro	90	74.7
mmmlu	90.8	81.4
MMMU	80.7	72.2
MMMU-Pro	74	59.9
MMStar	73.2	63.9
multichallenge	59	40.3
MultiNRC	48.6	12.4
OCRBench	86.5	806
OmniScience Accuracy	46.6	23.7
OmniScience Non-Hallucination	39	62.1
OSWorld-Verified	66.3	5
RealWorldQA	77	75.4
scicode	50	33.4
SEAL VISTA	46.4	34.9
simplebench	62	17.8
supergpqa	70.6	42.4
swe_bench_bash	76.8	21.6
SWE-bench Verified	81.5	38.8
Tau2 retail	88.9	63.4
Terminal-Bench Hard	47	8.3
vectara_answer_rate	98.7	93.8
vectara_avg_summary_length	114.5	86.6
vectara_factual_consistency	89.1	90.4
vectara_hallucination_rate ↓	10.9	9.6
VideoMMMU	84.4	61.2
τ²-Bench Telecom (AA run)	89.5	28.9

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.