Which is better, Claude Opus 4.6 or GPT-5.2?

Across 75 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 47 and GPT-5.2 leads on 28 — Claude Opus 4.6 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 4.6 vs GPT-5.2

AnthropicvsOpenAI75 shared benchmarks47–28 head-to-head

Benchmark	Claude Opus 4.6	GPT-5.2
AA Agentic Index	67.6	60.2
AA Intelligence	44.9	43.3
AA-LCR	74.3	79.3
AA-Omniscience	13.7	0.3
AIME 2025	99.8	100
aime_2026	96.7	98.3
AIME25 no tools	95.6	98
apexAgents	33	23
arc_agi_1	93	86.2
ARC-AGI-2	68.8	52.9
arena_elo	1497	1437
arena_text_factuality	1487	1442
arena_vision	1300	1245
Artificial Analysis Coding Index	48.1	48.7
browsecomp	86.8	65.8
browsecomp_with_context_manager	84	65.8
charxiv_rq	69.1	82.1
coding_arena_elo	1545	1418
critpt	12.6	11.6
deepsearchqa_f1	91.3	71.3
Fortress	20.5	17.5
frontiermath_tier_4	22.9	18.8
gdpval	55.9	48.3
GDPval-AA (Elo)	1619	1462
GPQA Diamond	91.3	92.4
HLE	53.1	45.5
HLE (with tools)	62.7	45.5
hmmt_feb_2026	96.2	97
hmmt_nov_2025	96.3	99.2
IFBench	62.5	75.4
imo_answer_bench	75.3	86.3
livebench	76.3	74.8
livebench_agentic_coding	49	50.3
livebench_coding	78.2	76.1
livebench_data_analysis	69.9	78.2
livebench_instruction_following	63.3	61.8
livebench_language	83.3	79.8
livebench_math	89.3	93.2
livebench_reasoning	88.7	83.2
LiveCodeBench	88.8	89
matharena_visual_math_overall	72.3	86.5
mathvision	71.2	83
MCP Atlas	76.8	68
metr_hcast	77	77
metr_re_bench	97.4	35
metr_swaa	98.5	99.6
MMLU-Pro	89.1	87
mmmlu	91.1	89.6
MMMU-Pro	77.3	79.5
mrcr_v2_8needle_128k_average	84	83.8
MultiNRC	57.1	42.2
ocrbench_v2	48.4	50.5
OmniDocBench 1.5	86.6	85.7
OmniScience Accuracy	47	44.3
OmniScience Non-Hallucination	37.2	38.4
scicode	52	52.1
screenspot_pro_no_tools	57.7	86.3
SEAL VISTA	46.1	46.6
simplebench	67.6	45.8
simpleqa_verified	46.5	38.9
swe_bench_bash	75.6	72.8
SWE-bench Multilingual	77.8	72
SWE-bench Pro	57.3	55.6
SWE-bench Verified	80.8	80
Tau2 retail	91.9	82
Terminal-Bench 2.0	65.4	54
Terminal-Bench Hard	65.4	62.2
tool_decathlon	47.2	46.3
toolathlon	56.8	46.3
vectara_answer_rate	99.8	100
vectara_avg_summary_length	137.6	186.3
vectara_factual_consistency	87.8	89.2
vectara_hallucination_rate ↓	12.2	10.8
τ²-Bench Telecom (AA run)	99.3	98.7
τ³-Bench	72.4	12

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.