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

Across 52 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.5 leads on 6 and Gemini 3.1 Pro leads on 46 — 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 Sonnet 4.5 vs Gemini 3.1 Pro

AnthropicvsGoogle52 shared benchmarks6–46 head-to-head

Benchmark	Claude Sonnet 4.5	Gemini 3.1 Pro
AA Agentic Index	50.6	23
AA Intelligence	37.4	47.7
AA-LCR	68.3	79
AA-Omniscience	-0.1	32.9
arc_agi_1	63.7	98
ARC-AGI-2	13.6	77.1
arena_elo	1456	1486
arena_text_factuality	1448	1471
Artificial Analysis Coding Index	52.1	68.8
browsecomp	24.1	85.9
charxiv_rq	67.2	89.9
coding_arena_elo	1392	1447
critpt	1.1	17.7
Fortress	17	29.8
frontiermath_tier_4	4.2	16.7
gdpval	40.4	23.3
GPQA Diamond	83.4	94.3
HLE	19.8	51.4
HLE (with tools)	32	51.6
hmmt_nov_2025	81.7	94.8
IFBench	57.3	77.1
imo_answer_bench	65.9	91
LiveCodeBench	71	91.7
LiveCodeBench v6	64	91.7
matharena_visual_math_overall	75.8	89.4
MCP Atlas	59.5	78.2
MMLU-Pro	88.2	91
mmmlu	89.1	92.6
MMMU-Pro	68.7	83
multichallenge	55.3	71.4
MultiNRC	35.8	64.7
OmniScience Accuracy	32.9	55.3
OmniScience Non-Hallucination	50.9	50.1
OSWorld-Verified	61.4	76.2
scicode	45	59
simplebench	54.3	79.6
simpleqa_verified	23.6	77.3
SWE-bench Multilingual	68	76.9
SWE-bench Pro	43.6	54.2
SWE-bench Verified	82	80.6
TauBench V3 - Banking	24.5	21.4
Terminal-Bench 2.0	50	68.5
Terminal-Bench 2.1	55.8	74
Terminal-Bench Hard	35.6	68.5
toolathlon	41	48.8
vectara_answer_rate	95.6	99.4
vectara_avg_summary_length	127.8	107.7
vectara_factual_consistency	88	89.6
vectara_hallucination_rate ↓	12	10.4
VisualToolBench	6.2	29
τ²-Bench Telecom (AA run)	78.1	99.3
τ³-Bench	19	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.