Which is better, Claude Opus 4.6 or Gemini 3.1 Pro?

Across 101 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 43 and Gemini 3.1 Pro leads on 57 — Gemini 3.1 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 4.6 vs Gemini 3.1 Pro

AnthropicvsGoogle101 shared benchmarks43–57 head-to-head

Benchmark	Claude Opus 4.6	Gemini 3.1 Pro
AA Agentic Index	67.6	23
AA Intelligence	44.9	47.7
AA-LCR	74.3	79
AA-Omniscience	13.7	32.9
AgentWorldBench - Android	61.7	61.4
AgentWorldBench - MCP	69.9	59.1
AgentWorldBench - OS	70.2	66.9
AgentWorldBench - Overall	57.8	54.6
AgentWorldBench - Search	29.3	30.2
AgentWorldBench - SWE	64.5	59.1
AgentWorldBench - Terminal	57.5	52.5
AgentWorldBench - Web	51.4	52.8
aime_2026	96.7	98.3
Apex (Pass@1)	34.5	60.9
Apex Shortlist (Pass@1)	85.9	89.1
apexAgents	33	33.5
arc_agi_1	93	98
arc_agi_3	0.5	0.4
ARC-AGI-2	68.8	77.1
arena_elo	1497	1486
arena_text_factuality	1487	1471
arena_vision	1300	1280
Artificial Analysis Coding Index	48.1	68.8
baby_vision	14.8	51.6
baby_vision_with_python	38.4	68.3
browsecomp	86.8	85.9
BrowseComp (Agent Swarm)	83.7	85.9
browsecomp_with_context_manager	84	85.9
charxiv_reasoning_with_tools	84.7	83.2
charxiv_rq	69.1	89.9
charxiv_rq_with_python	84.7	89.9
Chinese SimpleQA (C-SimpleQA)	76.4	85.9
coding_arena_elo	1545	1447
CorpusQA 1M (ACC)	71.7	53.8
critpt	12.6	17.7
cybergym	73.8	38.8
deepsearchqa_accuracy	80.6	60.2
deepsearchqa_f1	91.3	81.9
finance_agent	76.7	59.7
Fortress	20.5	29.8
frontiermath_tier_4	22.9	16.7
gdpval	55.9	23.3
GDPval-AA (Elo)	1619	1317
GPQA Diamond	91.3	94.3
HiL-Bench	38.3	35.3
HLE	53.1	51.4
HLE (with tools)	62.7	51.6
hmmt_feb_2026	96.2	94.7
hmmt_nov_2025	96.3	94.8
IFBench	62.5	77.1
imo_answer_bench	75.3	91
Legal Agent Benchmark	4.2	0
livebench	76.3	79.9
livebench_agentic_coding	49	44.1
livebench_coding	78.2	76.5
livebench_data_analysis	69.9	78.5
livebench_instruction_following	63.3	79.1
livebench_language	83.3	85.4
livebench_math	89.3	91
livebench_reasoning	88.7	84
LiveCodeBench	88.8	91.7
LiveCodeBench v6	88.8	91.7
matharena_visual_math_overall	72.3	89.4
mathvision	71.2	89.8
mathvision_with_python	84.6	95.7
MCP Atlas	76.8	78.2
MCPAtlas Public (Pass@1)	73.8	69.2
mcpmark	56.7	55.9
MMLU-Pro	89.1	91
mmmlu	91.1	92.6
mmmu_pro_with_python	77.3	85.3
MMMU-Pro	77.3	83
MRCR 1M (MMR)	92.9	76.3
mrcr_v2_8needle_128k_average	84	84.9
multichallenge	56	71.4
MultiNRC	57.1	64.7
nl2repo	49.8	33.4
officeqa_pro	57.1	42.9
OJBench (python)	60.3	70.7
OmniScience Accuracy	47	55.3
OmniScience Non-Hallucination	37.2	50.1
OSWorld-Verified	72.7	76.2
scicode	52	59
simplebench	67.6	79.6
simpleqa_verified	46.5	77.3
SWE-bench Multilingual	77.8	76.9
SWE-bench Pro	57.3	54.2
SWE-bench Verified	80.8	80.6
Terminal-Bench 2.0	65.4	68.5
Terminal-Bench Hard	65.4	68.5
tool_decathlon	47.2	48.8
toolathlon	56.8	48.8
usamo_2026	66.2	74.4
V* (w/ python)	86.4	96.9
vectara_answer_rate	99.8	99.4
vectara_avg_summary_length	137.6	107.7
vectara_factual_consistency	87.8	89.6
vectara_hallucination_rate ↓	12.2	10.4
VisualToolBench	27.5	29
τ²-Bench Telecom (AA run)	99.3	99.3
τ³-Bench	72.4	67.1

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.