Which is better, Claude Opus 4.5 or GPT-5?

Across 58 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 33 and GPT-5 leads on 24 — 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-5

AnthropicvsOpenAI58 shared benchmarks33–24 head-to-head

Benchmark	Claude Opus 4.5	GPT-5
AA Agentic Index	59.6	49.7
AA Intelligence	41.9	35.3
AA-LCR	76	76.3
AA-Omniscience	14	-8.7
AIME 2025	92.8	95
AIME25 no tools	91	94.6
arc_agi_1	82	56.2
ARC-AGI-2	37.6	7.5
Artificial Analysis Coding Index	47.8	38.9
browsecomp	37	54.9
browsecomp_zh	62.4	65
coding_arena_elo	1495	1419
critpt	4.6	5.7
ERQA	46.8	65.7
Fortress	13.6	17
frontiermath_tier_4	4.2	12.5
gdpval	47.3	32.5
GPQA Diamond	87	87.3
HLE	30.8	28.5
HLE (with tools)	43.4	35.2
hmmt_feb_2025	92.9	88.3
hmmt_nov_2025	93.3	89.2
IFBench	58	73.1
imo_answer_bench	84	76
LiveCodeBench	87	85
LiveCodeBench v6	84.8	87
metr_hcast	73.4	68.4
metr_re_bench	40	40
metr_swaa	99.5	100
mmlu_redux	95.6	95.3
MMLU-Pro	90	87.5
MMMU	80.7	84.2
MMMU-Pro	74	78.4
Multi-SWE-Bench	50	39.3
multichallenge	59	71.1
MultiNRC	48.6	52.1
OJ-Bench (cpp)	54.6	56.2
OmniScience Accuracy	46.6	40.3
OmniScience Non-Hallucination	39	21
scicode	50	43
SEAL VISTA	46.4	49.7
Seal-0	47.7	51.4
simplebench	62	56.7
simpleqa_verified	41.8	50.6
swe_bench_bash	76.8	65
SWE-bench Multilingual	77.5	55.3
SWE-bench Pro	57.1	41.8
SWE-bench Verified	81.5	74.9
Tau2 retail	88.9	81.1
Terminal-Bench 2.0	59.3	35.2
Terminal-Bench Hard	47	37.9
vectara_answer_rate	98.7	99.9
vectara_avg_summary_length	114.5	162.7
vectara_factual_consistency	89.1	85.3
vectara_hallucination_rate ↓	10.9	14.7
VideoMMMU	84.4	84.6
τ²-Bench	98.2	85
τ²-Bench Telecom (AA run)	89.5	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.