Which is better, Claude Opus 4.5 or GPT-5.2?

Across 87 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 35 and GPT-5.2 leads on 52 — GPT-5.2 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-5.2

AnthropicvsOpenAI87 shared benchmarks35–52 head-to-head

Benchmark	Claude Opus 4.5	GPT-5.2
AA Agentic Index	59.6	60.2
AA Intelligence	41.9	43.3
AA-LCR	76	79.3
AA-Omniscience	14	0.3
AIME 2025	92.8	100
aime_2026	95.1	98.3
AIME25 no tools	91	98
arc_agi_1	82	86.2
ARC-AGI-2	37.6	52.9
arena_elo	1473	1437
arena_text_factuality	1458	1442
Artificial Analysis Coding Index	47.8	48.7
browsecomp	37	65.8
BrowseComp (context management)	59.2	70
browsecomp_with_context_manager	67.8	65.8
browsecomp_zh	62.4	76.1
charxiv_rq	68.5	82.1
coding_arena_elo	1495	1418
critpt	4.6	11.6
deepsearchqa_f1	76.1	71.3
Fortress	13.6	17.5
frontiermath_tier_4	4.2	18.8
gdpval	47.3	48.3
GPQA Diamond	87	92.4
HLE	30.8	45.5
HLE (with tools)	43.4	45.5
hmmt_feb_2025	92.9	100
hmmt_feb_2026	85.3	97
hmmt_nov_2025	93.3	99.2
IFBench	58	75.4
imo_answer_bench	84	86.3
InfoVQA (val)	76.9	84
livebench	76	74.8
livebench_agentic_coding	39.7	50.3
livebench_coding	79.7	76.1
livebench_data_analysis	74.4	78.2
livebench_instruction_following	62.5	61.8
livebench_language	81.3	79.8
livebench_math	90.4	93.2
livebench_reasoning	80.1	83.2
LiveCodeBench	87	89
longbench_v2	64.4	54.5
LongVideoBench	67.2	76.5
mathvision	77.1	83
MathVista	80.2	82.8
MCP Atlas	69.8	68
metr_hcast	73.4	77
metr_re_bench	40	35
metr_swaa	99.5	99.6
MMLU-Pro	90	87
mmmlu	90.8	89.6
MMMU-Pro	74	79.5
MMVU	77.3	80.8
MotionBench	60.3	64.8
MultiNRC	48.6	42.2
OCRBench	86.5	80.7
OmniDocBench 1.5	87.7	85.7
OmniScience Accuracy	46.6	44.3
OmniScience Non-Hallucination	39	38.4
PaperBench	72.9	63.7
scicode	50	52.1
SEAL VISTA	46.4	46.6
Seal-0	47.7	45
simplebench	62	45.8
simpleqa_verified	41.8	38.9
SimpleVQA	69.7	55.8
swe_bench_bash	76.8	72.8
SWE-bench Multilingual	77.5	72
SWE-bench Pro	57.1	55.6
SWE-bench Verified	81.5	80
SWE-Perf	4.7	3.6
SWT-bench	80.2	80.7
Tau2 retail	88.9	82
Terminal-Bench 2.0	59.3	54
Terminal-Bench Hard	47	62.2
tool_decathlon	43.5	46.3
toolathlon	43.5	46.3
vectara_answer_rate	98.7	100
vectara_avg_summary_length	114.5	186.3
vectara_factual_consistency	89.1	89.2
vectara_hallucination_rate ↓	10.9	10.8
VideoMMMU	84.4	85.9
WorldVQA	36.8	28
ZeroBench	3	9
ZeroBench (w/ tools)	9	7
τ²-Bench	98.2	85.5
τ²-Bench Telecom (AA run)	89.5	98.7

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.