Which is better, Claude Opus 4.5 or GPT-5.5?

Across 52 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 4 and GPT-5.5 leads on 48 — GPT-5.5 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-5.5

AnthropicvsOpenAI52 shared benchmarks4–48 head-to-head

Benchmark	Claude Opus 4.5	GPT-5.5
AA Agentic Index	59.6	47.4
AA Intelligence	41.9	56.3
AA-LCR	76	79
AA-Omniscience	14	20.5
aime_2026	95.1	100
arc_agi_1	82	95
ARC-AGI-2	37.6	85
arena_elo	1473	1482
arena_text_factuality	1458	1482
Artificial Analysis Coding Index	47.8	74.9
browsecomp	37	84.4
coding_arena_elo	1495	1457
critpt	4.6	27.1
cybergym	50.6	81.8
finance_agent	55.2	60
Fortress	13.6	16.3
frontiermath_tier_4	4.2	35.4
gdpval	47.3	49.5
GPQA Diamond	87	93.6
HLE	30.8	52.2
HLE (with tools)	43.4	52.2
hmmt_feb_2026	85.3	98.5
hmmt_nov_2025	93.3	96.5
IFBench	58	75.9
livebench	76	80.7
livebench_agentic_coding	39.7	54
livebench_coding	79.7	82.2
livebench_data_analysis	74.4	81.6
livebench_instruction_following	62.5	70.7
livebench_language	81.3	87.4
livebench_math	90.4	95.9
livebench_reasoning	80.1	89.7
MCP Atlas	69.8	82.8
MMMU-Pro	74	83.2
nl2repo	43.2	50.7
OmniScience Accuracy	46.6	58
OmniScience Non-Hallucination	39	12
OSWorld-Verified	66.3	78.7
scicode	50	56.1
simplebench	62	69
simpleqa_verified	41.8	63.1
SWE-bench Pro	57.1	58.6
SWE-bench Verified	81.5	80.6
Terminal-Bench 2.0	59.3	82.7
Terminal-Bench Hard	47	60.6
tool_decathlon	43.5	55.6
toolathlon	43.5	55.6
vectara_answer_rate	98.7	100
vectara_avg_summary_length	114.5	129.6
vectara_factual_consistency	89.1	90.7
vectara_hallucination_rate ↓	10.9	9.3
τ²-Bench Telecom (AA run)	89.5	93.9

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.