Which is better, Claude Opus 4.5 or GPT-5.4?

Across 59 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 10 and GPT-5.4 leads on 49 — GPT-5.4 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-5.4

AnthropicvsOpenAI59 shared benchmarks10–49 head-to-head

Benchmark	Claude Opus 4.5	GPT-5.4
AA Agentic Index	59.6	58.2
AA Intelligence	41.9	53.1
AA-LCR	76	77.7
AA-Omniscience	14	5.8
aime_2026	95.1	99.2
arc_agi_1	82	93.7
ARC-AGI-2	37.6	74
arena_elo	1473	1477
arena_text_factuality	1458	1476
Artificial Analysis Coding Index	47.8	71.1
browsecomp	37	82.7
browsecomp_with_context_manager	67.8	82.7
charxiv_rq	68.5	82.8
coding_arena_elo	1495	1457
critpt	4.6	23.4
cybergym	50.6	66.3
deepsearchqa_f1	76.1	78.6
finance_agent	55.2	57.2
frontiermath_tier_4	4.2	27.1
gdpval	47.3	50.1
GPQA Diamond	87	93
HLE	30.8	43.7
HLE (with tools)	43.4	52.1
hmmt_feb_2026	85.3	97.7
hmmt_nov_2025	93.3	95.8
IFBench	58	73.9
imo_answer_bench	84	91.4
livebench	76	80.3
livebench_agentic_coding	39.7	53.8
livebench_coding	79.7	77.5
livebench_data_analysis	74.4	79.3
livebench_instruction_following	62.5	70.2
livebench_language	81.3	82.6
livebench_math	90.4	94.2
livebench_reasoning	80.1	88.1
mathvision	77.1	92
MCP Atlas	69.8	70.6
MMLU-Pro	90	87.5
MMMU-Pro	74	81.2
MultiNRC	48.6	58.3
nl2repo	43.2	41.3
OmniDocBench 1.5	87.7	89.1
OmniScience Accuracy	46.6	50.9
OmniScience Non-Hallucination	39	17.4
OSWorld-Verified	66.3	75
scicode	50	56.6
SEAL VISTA	46.4	50.9
simpleqa_verified	41.8	45.3
SWE-bench Multilingual	77.5	71.7
SWE-bench Pro	57.1	59.1
Terminal-Bench 2.0	59.3	75.1
Terminal-Bench Hard	47	57.6
tool_decathlon	43.5	54.6
toolathlon	43.5	54.6
vectara_answer_rate	98.7	99.9
vectara_avg_summary_length	114.5	81.7
vectara_factual_consistency	89.1	93
vectara_hallucination_rate ↓	10.9	7
τ²-Bench Telecom (AA run)	89.5	87.1

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.