Which is better, Claude Sonnet 4 or Gemini 3 Pro?

Across 46 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 6 and Gemini 3 Pro leads on 40 — Gemini 3 Pro 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 Gemini 3 Pro

AnthropicvsGoogle46 shared benchmarks6–40 head-to-head

Benchmark	Claude Sonnet 4	Gemini 3 Pro
AA Agentic Index	39.2	52
AA Intelligence	29.8	40.6
AA-LCR	69.7	73
AA-Omniscience	0.2	15.3
AIME 2025	74	100
arc_agi_1	63.7	75
ARC-AGI-2	13.6	54
arena_vision	1207	1290
Artificial Analysis Coding Index	37.6	46.5
browsecomp	12.2	59.2
browsecomp_zh	29.1	66.8
critpt	1.1	9.1
Fortress	24.4	41.7
frontiermath_tier_4	0	18.8
gdpval	31.2	34.2
GPQA Diamond	78	91.9
HLE	10.7	45.8
HLE (with tools)	20.3	45.8
IFBench	55	70.4
livebench	74.8	73.4
LiveCodeBench	68.5	92
LiveCodeBench v6	48.5	90.7
MMLU-Pro	84	90.1
mmmlu	86.5	91.8
MMMU-Pro	62.4	81
Multi-SWE-Bench	35.7	42.7
multichallenge	57.1	65.7
MultiNRC	18.4	59
OmniScience Accuracy	22.7	55.8
OmniScience Non-Hallucination	70.9	10
scicode	40	56.1
SEAL VISTA	45.5	51.5
simplebench	45.5	76.4
simpleqa	15.9	72.1
swe_bench_bash	64.9	74.2
SWE-bench Multilingual	56.9	68.7
SWE-bench Pro	42.7	43.3
SWE-bench Verified	80.2	78
Terminal-Bench Hard	31.1	56.9
vectara_answer_rate	98.6	99.4
vectara_avg_summary_length	145.8	101.9
vectara_factual_consistency	89.7	86.4
vectara_hallucination_rate ↓	10.3	13.6
VisualToolBench	4.5	26.9
τ²-Bench	65	98
τ²-Bench Telecom (AA run)	64.6	98

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.