Which is better, Claude Opus 4.6 or Gemini 3 Pro?

Across 71 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 45 and Gemini 3 Pro leads on 26 — Claude Opus 4.6 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 4.6 vs Gemini 3 Pro

AnthropicvsGoogle71 shared benchmarks45–26 head-to-head

Benchmark	Claude Opus 4.6	Gemini 3 Pro
AA Agentic Index	67.6	52
AA Intelligence	44.9	40.6
AA-LCR	74.3	73
AA-Omniscience	13.7	15.3
AIME 2025	99.8	100
aime_2026	96.7	91.7
AIME25 no tools	95.6	96
apexAgents	33	18.4
arc_agi_1	93	75
ARC-AGI-2	68.8	54
arena_elo	1497	1485
arena_text_factuality	1487	1481
arena_vision	1300	1290
Artificial Analysis Coding Index	48.1	46.5
browsecomp	86.8	59.2
browsecomp_with_context_manager	84	59.2
charxiv_rq	69.1	81.4
coding_arena_elo	1545	1438
critpt	12.6	9.1
cybergym	73.8	39.9
deepsearchqa_f1	91.3	63.2
Fortress	20.5	41.7
frontiermath_tier_4	22.9	18.8
gdpval	55.9	34.2
GDPval-AA (Elo)	1619	1195
GPQA Diamond	91.3	91.9
HLE	53.1	45.8
HLE (with tools)	62.7	45.8
hmmt_feb_2026	96.2	86.4
hmmt_nov_2025	96.3	93.3
IFBench	62.5	70.4
imo_answer_bench	75.3	83.3
livebench	76.3	73.4
LiveCodeBench	88.8	92
LiveCodeBench v6	88.8	90.7
matharena_visual_math_overall	72.3	84.2
mathvision	71.2	86.1
MCP Atlas	76.8	70.3
metr_hcast	77	71.3
metr_re_bench	97.4	35
metr_swaa	98.5	99.5
MMLU-Pro	89.1	90.1
mmmlu	91.1	91.8
MMMU-Pro	77.3	81
mrcr_v2_8needle_128k_average	84	77
multichallenge	56	65.7
MultiNRC	57.1	59
ocrbench_v2	48.4	63.4
OmniDocBench 1.5	86.6	88.5
OmniScience Accuracy	47	55.8
OmniScience Non-Hallucination	37.2	10
scicode	52	56.1
screenspot_pro_no_tools	57.7	72.7
SEAL VISTA	46.1	51.5
simplebench	67.6	76.4
simpleqa_verified	46.5	72.9
swe_bench_bash	75.6	74.2
SWE-bench Multilingual	77.8	68.7
SWE-bench Pro	57.3	43.3
SWE-bench Verified	80.8	78
Terminal-Bench 2.0	65.4	54.2
Terminal-Bench Hard	65.4	56.9
tool_decathlon	47.2	36.4
toolathlon	56.8	36.4
vectara_answer_rate	99.8	99.4
vectara_avg_summary_length	137.6	101.9
vectara_factual_consistency	87.8	86.4
vectara_hallucination_rate ↓	12.2	13.6
Vending-Bench 2	801759	547816
VisualToolBench	27.5	26.9
τ²-Bench Telecom (AA run)	99.3	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.