Which is better, Claude Opus 4.5 or Gemini 3 Pro?

Across 97 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 42 and Gemini 3 Pro leads on 53 — 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 Opus 4.5 vs Gemini 3 Pro

AnthropicvsGoogle97 shared benchmarks42–53 head-to-head

Benchmark	Claude Opus 4.5	Gemini 3 Pro
AA Agentic Index	59.6	52
AA Intelligence	41.9	40.6
AA-LCR	76	73
AA-Omniscience	14	15.3
AIME 2025	92.8	100
AIME 2026 I	93.3	90.6
aime_2026	95.1	91.7
AIME25 no tools	91	96
arc_agi_1	82	75
ARC-AGI-2	37.6	54
ARC-AGI-2 (Verified)	37.6	31.1
arena_elo	1473	1485
arena_text_factuality	1458	1481
Artificial Analysis Coding Index	47.8	46.5
browsecomp	37	59.2
BrowseComp (context management)	59.2	67.6
browsecomp_with_context_manager	67.8	59.2
browsecomp_zh	62.4	66.8
charxiv_rq	68.5	81.4
coding_arena_elo	1495	1438
critpt	4.6	9.1
cybergym	50.6	39.9
deepsearchqa_f1	76.1	63.2
FinSearchCompT2&T3	66.2	49.9
Fortress	13.6	41.7
frontiermath_tier_4	4.2	18.8
gdpval	47.3	34.2
GPQA Diamond	87	91.9
HLE	30.8	45.8
HLE (with tools)	43.4	45.8
hmmt_feb_2025	92.9	97.5
hmmt_feb_2026	85.3	86.4
hmmt_nov_2025	93.3	93.3
IFBench	58	70.4
imo_answer_bench	84	83.3
InfoVQA (val)	76.9	57.2
livebench	76	73.4
LiveCodeBench	87	92
LiveCodeBench v6	84.8	90.7
longbench_v2	64.4	68.2
LongVideoBench	67.2	77.7
mathvision	77.1	86.1
MathVista	80.2	89.8
MCP Atlas	69.8	70.3
metr_hcast	73.4	71.3
metr_re_bench	40	35
metr_swaa	99.5	99.5
MLVU	81.7	80.7
MMLU-Pro	90	90.1
mmmlu	90.8	91.8
MMMU-Pro	74	81
MMVU	77.3	78.9
MotionBench	60.3	70.3
Multi-SWE-Bench	50	42.7
multichallenge	59	65.7
MultiNRC	48.6	59
OCRBench	86.5	90.3
OctoCodingbench	36.2	22.9
OJ-Bench (cpp)	54.6	68.5
OJBench (cpp)	54.6	68.5
OmniDocBench 1.5	87.7	88.5
OmniScience Accuracy	46.6	55.8
OmniScience Non-Hallucination	39	10
scicode	50	56.1
SEAL VISTA	46.4	51.5
Seal-0	47.7	45.5
simplebench	62	76.4
simpleqa_verified	41.8	72.9
SimpleVQA	69.7	69.7
swe_bench_bash	76.8	74.2
SWE-bench Multilingual	77.5	68.7
SWE-bench Pro	57.1	43.3
SWE-bench Verified	81.5	78
SWE-Perf	4.7	6.5
SWT-bench	80.2	79.7
Terminal-Bench 2.0	59.3	54.2
Terminal-Bench Hard	47	56.9
tool_decathlon	43.5	36.4
toolathlon	43.5	36.4
vectara_answer_rate	98.7	99.4
vectara_avg_summary_length	114.5	101.9
vectara_factual_consistency	89.1	86.4
vectara_hallucination_rate ↓	10.9	13.6
VIBE (Average)	90.7	82.4
VIBE-Android	92.2	78.7
VIBE-Backend	98	78.7
VIBE-iOS	90	75.8
VIBE-Simulation	84	89.2
VIBE-Web	89.1	89.5
VideoMMMU	84.4	87.6
WideSearch (item-f1)	76.2	57
WorldVQA	36.8	47.4
ZeroBench	3	8
ZeroBench (w/ tools)	9	12
τ²-Bench	98.2	98
τ²-Bench (Retail)	88.9	85.3
τ²-Bench Telecom (AA run)	89.5	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.