Which is better, Claude Opus 4.7 or GPT-5.2?

Across 55 shared benchmarks tracked on Vector Wire, Claude Opus 4.7 leads on 38 and GPT-5.2 leads on 16 — Claude Opus 4.7 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 4.7 vs GPT-5.2

AnthropicvsOpenAI55 shared benchmarks38–16 head-to-head

Benchmark	Claude Opus 4.7	GPT-5.2
AA Agentic Index	64.6	60.2
AA Intelligence	55	43.3
AA-LCR	75.3	79.3
AA-Omniscience	27.3	0.3
aime_2026	95.8	98.3
arc_agi_1	92	86.2
ARC-AGI-2	75.8	52.9
arena_elo	1494	1437
arena_text_factuality	1474	1442
arena_vision	1304	1245
Artificial Analysis Coding Index	73.6	48.7
browsecomp	79.8	65.8
charxiv_rq	82.1	82.1
coding_arena_elo	1557	1418
critpt	12	11.6
frontiermath_tier_4	22.9	18.8
gaia_overall	28	77.4
gdpval	58.6	48.3
GDPval-AA (Elo)	1753	1462
GPQA Diamond	94.2	92.4
HLE	54.7	45.5
HLE (with tools)	54.7	45.5
hmmt_feb_2026	93.9	97
IFBench	58.6	75.4
livebench	76.9	74.8
livebench_agentic_coding	50.7	50.3
livebench_coding	82.1	76.1
livebench_data_analysis	78.3	78.2
livebench_instruction_following	66.7	61.8
livebench_language	77.9	79.8
livebench_math	92.8	93.2
livebench_reasoning	87.2	83.2
MCP Atlas	79.1	68
mmmlu	91.5	89.6
MMMU-Pro	78.8	79.5
mrcr_v2_8needle_128k_average	59.3	83.8
OmniScience Accuracy	48.9	44.3
OmniScience Non-Hallucination	57.7	38.4
scicode	54.5	52.1
screenspot_pro_no_tools	79.5	86.3
simplebench	61.7	45.8
simpleqa_verified	50.6	38.9
SWE-bench Multilingual	80.5	72
SWE-bench Pro	64.3	55.6
SWE-bench Verified	87.6	80
TauBench V3 - Banking	34.6	11.1
Terminal-Bench 2.0	69.4	54
Terminal-Bench Hard	54.5	62.2
toolathlon	59.3	46.3
vectara_answer_rate	98	100
vectara_avg_summary_length	149.1	186.3
vectara_factual_consistency	88	89.2
vectara_hallucination_rate ↓	12	10.8
τ²-Bench Telecom (AA run)	88.6	98.7
τ³-Bench	28.9	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.