Which is better, Claude Sonnet 4 or o4-mini?

Across 48 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 27 and o4-mini leads on 19 — 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 o4-mini

AnthropicvsOpenAI48 shared benchmarks27–19 head-to-head

Benchmark	Claude Sonnet 4	o4-mini
AA Agentic Index	39.2	36.1
AA Intelligence	29.8	26.1
AA-LCR	69.7	60
AA-Omniscience	0.2	-35.7
aider_polyglot	70.7	72
AIME 2025	74	92.7
AIR-Bench 2024	88.3	78.5
anthropic_red_team	99.2	98.2
arc_agi_1	63.7	58.7
ARC-AGI-2	13.6	6.1
arena_vision	1207	1201
Artificial Analysis Coding Index	37.6	25.6
bbq	97.9	94
browsecomp	12.2	51.5
critpt	1.1	0.6
Fortress	24.4	21.5
frontiermath_tier_4	0	6.3
gdpval	31.2	25.4
GPQA Diamond	78	81.4
harmbench	98	97
HLE	10.7	16.5
IFBench	55	68.7
LiveCodeBench	68.5	84.5
livecodebench_easy	98.8	98.8
livecodebench_hard	32.6	62.9
livecodebench_medium	76	92.2
MMMU	74.4	81.6
MMMU-Pro	62.4	69.2
multichallenge	57.1	44.9
MultiNRC	18.4	22.2
OmniScience Accuracy	22.7	24.8
OmniScience Non-Hallucination	70.9	19.5
scicode	40	46.5
SEAL VISTA	45.5	51.8
simple_safety_tests	100	100
simplebench	45.5	38.7
swe_bench_bash	64.9	45
SWE-bench Verified	80.2	68.1
TAU-bench (airline)	60	49.2
TAU-bench (retail)	80.5	71.8
Terminal-Bench Hard	31.1	15.2
vectara_answer_rate	98.6	99.2
vectara_avg_summary_length	145.8	130.9
vectara_factual_consistency	89.7	81.4
vectara_hallucination_rate ↓	10.3	18.6
VisualToolBench	4.5	11.1
xstest	97.2	97.4
τ²-Bench Telecom (AA run)	64.6	55.6

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.