Which is better, Claude Sonnet 4.5 or Gemini 2.5 Pro?

Across 57 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 36 and Gemini 2.5 Pro leads on 20 — Claude Sonnet 4.5 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 4.5 vs Gemini 2.5 Pro

AnthropicvsGoogle57 shared benchmarks36–20 head-to-head

Benchmark	Claude Sonnet 4.5	Gemini 2.5 Pro
AA Agentic Index	50.6	7.2
AA Intelligence	37.4	27
AA-LCR	68.3	66
AA-Omniscience	-0.1	-14.3
AIME 2025	88	88.3
AIR-Bench 2024	89.8	73.6
arc_agi_1	63.7	41
ARC-AGI-2	13.6	4.9
arena_elo	1456	1446
arena_text_factuality	1448	1462
ArtifactsBench	61.5	57.7
Artificial Analysis Coding Index	52.1	46.7
browsecomp	24.1	9.9
browsecomp_zh	42.4	32.2
coding_arena_elo	1392	1224
critpt	1.1	2.6
FinSearchComp-global	60.8	42.6
Fortress	17	54.9
frontiermath_tier_4	4.2	4.2
GAIA (text only)	71.2	60.2
gdpval	40.4	8.5
GPQA Diamond	83.4	84.4
HLE	19.8	22.5
HLE (with tools)	32	28.4
HMMT 2025	74.6	65.7
hmmt_feb_2025	79.2	82.5
hmmt_nov_2025	81.7	80
IFBench	57.3	49
LiveCodeBench	71	82.7
matharena_visual_math_overall	75.8	77.2
MathVista	79.8	83.9
MMLU-Pro	88.2	86
MMMU	79.6	83.9
MMMU-Pro	68.7	74.9
mrcr	55.4	93
multichallenge	55.3	53.6
OmniScience Accuracy	32.9	39
OmniScience Non-Hallucination	50.9	12.6
scicode	45	43
SEAL VISTA	48.8	50.8
simplebench	54.3	62.4
simpleqa_verified	23.6	56
swe_bench_bash	71.4	53.6
SWE-bench Verified	82	67.2
TauBench V3 - Banking	24.5	9.7
Terminal-Bench	51	25.3
Terminal-Bench 2.1	55.8	28.5
Terminal-Bench Hard	35.6	26.5
theagentcompany	41	39.3
vectara_answer_rate	95.6	99.1
vectara_avg_summary_length	127.8	106.4
vectara_factual_consistency	88	93
vectara_hallucination_rate ↓	12	7
xbench_deepsearch	66	56
τ²-Bench	98	54
τ²-Bench Telecom (AA run)	78.1	54.1
τ³-Bench	19	9.3

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.