Which is better, Claude Opus 4.5 or Gemini 3.1 Pro?

Across 64 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 10 and Gemini 3.1 Pro leads on 54 — 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 4.5 vs Gemini 3.1 Pro

AnthropicvsGoogle64 shared benchmarks10–54 head-to-head

Benchmark	Claude Opus 4.5	Gemini 3.1 Pro
AA Agentic Index	59.6	23
AA Intelligence	41.9	47.7
AA-LCR	76	79
AA-Omniscience	14	32.9
aime_2026	95.1	98.3
arc_agi_1	82	98
ARC-AGI-2	37.6	77.1
arena_elo	1473	1486
arena_text_factuality	1458	1471
Artificial Analysis Coding Index	47.8	68.8
browsecomp	37	85.9
browsecomp_with_context_manager	67.8	85.9
charxiv_rq	68.5	89.9
coding_arena_elo	1495	1447
critpt	4.6	17.7
cybergym	50.6	38.8
deepsearchqa_f1	76.1	81.9
finance_agent	55.2	59.7
Fortress	13.6	29.8
frontiermath_tier_4	4.2	16.7
gdpval	47.3	23.3
GPQA Diamond	87	94.3
HLE	30.8	51.4
HLE (with tools)	43.4	51.6
hmmt_feb_2026	85.3	94.7
hmmt_nov_2025	93.3	94.8
IFBench	58	77.1
imo_answer_bench	84	91
livebench	76	79.9
livebench_agentic_coding	39.7	44.1
livebench_coding	79.7	76.5
livebench_data_analysis	74.4	78.5
livebench_instruction_following	62.5	79.1
livebench_language	81.3	85.4
livebench_math	90.4	91
livebench_reasoning	80.1	84
LiveCodeBench	87	91.7
LiveCodeBench v6	84.8	91.7
mathvision	77.1	89.8
MCP Atlas	69.8	78.2
MMLU-Pro	90	91
mmmlu	90.8	92.6
MMMU-Pro	74	83
multichallenge	59	71.4
MultiNRC	48.6	64.7
nl2repo	43.2	33.4
OmniScience Accuracy	46.6	55.3
OmniScience Non-Hallucination	39	50.1
OSWorld-Verified	66.3	76.2
scicode	50	59
simplebench	62	79.6
simpleqa_verified	41.8	77.3
SWE-bench Multilingual	77.5	76.9
SWE-bench Pro	57.1	54.2
SWE-bench Verified	81.5	80.6
Terminal-Bench 2.0	59.3	68.5
Terminal-Bench Hard	47	68.5
tool_decathlon	43.5	48.8
toolathlon	43.5	48.8
vectara_answer_rate	98.7	99.4
vectara_avg_summary_length	114.5	107.7
vectara_factual_consistency	89.1	89.6
vectara_hallucination_rate ↓	10.9	10.4
τ²-Bench Telecom (AA run)	89.5	99.3

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.