Which is better, Claude Sonnet 4 or GPT-4o?

Across 59 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 53 and GPT-4o leads on 6 — Claude Sonnet 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 vs GPT-4o

AnthropicvsOpenAI59 shared benchmarks53–6 head-to-head

Benchmark	Claude Sonnet 4	GPT-4o
AA Agentic Index	39.2	9.7
AA Intelligence	29.8	11.2
AA-LCR	69.7	55
AA-Omniscience	0.2	-10.5
aider_polyglot	70.7	30.7
AIME 2024	43.4	9.3
AIME 2025	74	11.7
AIR-Bench 2024	88.3	62.4
anthropic_red_team	99.2	99.1
arc_agi_1	63.7	4.5
ARC-AGI-2	13.6	0
arena_vision	1207	1162
Artificial Analysis Coding Index	37.6	24.2
bbq	97.9	95.1
CNMO 2024	60.4	10.8
critpt	1.1	0
Fortress	24.4	47.2
gdpval	31.2	0
GPQA Diamond	78	70.1
harmbench	98	82.9
HLE	10.7	5.3
IFBench	55	36
ifeval	87.6	84.3
LiveCodeBench	68.5	38.3
LiveCodeBench v6	48.5	30.9
livecodebench_easy	98.8	82.5
livecodebench_hard	32.6	4.5
livecodebench_medium	76	32.1
math_500_em	94	74.6
mmlu	91.5	88.1
mmlu_redux	93.6	88
MMLU-Pro	84	74.7
mmmlu	86.5	81.4
MMMU	74.4	72.2
MMMU-Pro	62.4	59.9
multichallenge	57.1	40.3
MultiNRC	18.4	12.4
OmniScience Accuracy	22.7	23.7
OmniScience Non-Hallucination	70.9	62.1
OSWorld-Verified	42.2	5
scicode	40	33.4
SEAL VISTA	45.5	34.9
simple_safety_tests	100	98.5
simplebench	45.5	17.8
simpleqa	15.9	39.4
supergpqa	55.7	42.4
swe_bench_bash	64.9	21.6
SWE-bench Verified	80.2	38.8
TAU-bench (airline)	60	42.8
TAU-bench (retail)	80.5	60.3
Tau2 airline	55.5	45.5
Tau2 retail	75	63.4
Terminal-Bench Hard	31.1	8.3
vectara_answer_rate	98.6	93.8
vectara_avg_summary_length	145.8	86.6
vectara_factual_consistency	89.7	90.4
vectara_hallucination_rate ↓	10.3	9.6
xstest	97.2	97.3
τ²-Bench Telecom (AA run)	64.6	28.9

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.