Which is better, Claude Sonnet 4.6 or GPT-5.4?

Across 64 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 19 and GPT-5.4 leads on 44 — 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 Sonnet 4.6 vs GPT-5.4

AnthropicvsOpenAI64 shared benchmarks19–44 head-to-head

Benchmark	Claude Sonnet 4.6	GPT-5.4
AA Agentic Index	61.6	58.2
AA Intelligence	48.4	53.1
AA-LCR	74	77.7
AA-Omniscience	12.2	5.8
AgentWorldBench - Android	58	60
AgentWorldBench - MCP	70	70.1
AgentWorldBench - OS	63.2	68.6
AgentWorldBench - Overall	56	58.3
AgentWorldBench - Search	28.8	37.3
AgentWorldBench - SWE	64.5	66.3
AgentWorldBench - Terminal	57	53.7
AgentWorldBench - Web	50.8	51.8
apexAgents	28	33.3
arc_agi_1	86.5	93.7
ARC-AGI-2	60.4	74
arena_elo	1472	1477
arena_text_factuality	1460	1476
arena_vision	1278	1277
Artificial Analysis Coding Index	63	71.1
browsecomp	76.2	82.7
charxiv_rq	72.4	82.8
coding_arena_elo	1523	1457
critpt	3.1	23.4
DeepSWE 1.1	30	52
finance_agent	63.3	57.2
frontiermath_tier_4	8.3	27.1
gdpval	54.8	50.1
GDPval-AA (Elo)	1676	1674
GPQA Diamond	89.9	93
HLE	49	43.7
HLE (with tools)	46.8	52.1
IFBench	56.6	73.9
itbenchSre	39.8	34.5
Legal Agent Benchmark	5.4	0.4
livebench	75.5	80.3
livebench_agentic_coding	42.6	53.8
livebench_coding	79.3	77.5
livebench_data_analysis	78	79.3
livebench_instruction_following	63.2	70.2
livebench_language	76.1	82.6
livebench_math	87	94.2
livebench_reasoning	84.8	88.1
MCP Atlas	69.5	70.6
MMMU-Pro	75.6	81.2
officeqa	68.7	68.1
officeqa_pro	53.4	51.1
OmniScience Accuracy	40.9	50.9
OmniScience Non-Hallucination	51.6	17.4
OSWorld-Verified	78.5	75
scicode	47	56.6
simpleqa_verified	29	45.3
SWE-bench Pro	58.1	59.1
TauBench V3 - Banking	34.4	39.6
Terminal-Bench 2.0	59.1	75.1
Terminal-Bench 2.1	71.2	78.3
Terminal-Bench Hard	59.1	57.6
toolathlon	49.4	54.6
usamo_2026	55	95.2
vectara_answer_rate	99.9	99.9
vectara_avg_summary_length	114.7	81.7
vectara_factual_consistency	89.4	93
vectara_hallucination_rate ↓	10.6	7
τ²-Bench Telecom (AA run)	97.9	87.1
τ³-Bench	30.5	72.9

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.