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

Across 71 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 26 and Gemini 3.1 Pro leads on 44 — 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 4.6 vs Gemini 3.1 Pro

AnthropicvsGoogle71 shared benchmarks26–44 head-to-head

Benchmark	Claude Sonnet 4.6	Gemini 3.1 Pro
AA Agentic Index	61.6	23
AA Intelligence	48.4	47.7
AA-LCR	74	79
AA-Omniscience	12.2	32.9
AgentWorldBench - Android	58	61.4
AgentWorldBench - MCP	70	59.1
AgentWorldBench - OS	63.2	66.9
AgentWorldBench - Overall	56	54.6
AgentWorldBench - Search	28.8	30.2
AgentWorldBench - SWE	64.5	59.1
AgentWorldBench - Terminal	57	52.5
AgentWorldBench - Web	50.8	52.8
apexAgents	28	33.5
arc_agi_1	86.5	98
ARC-AGI-2	60.4	77.1
arena_elo	1472	1486
arena_text_factuality	1460	1471
arena_vision	1278	1280
Artificial Analysis Coding Index	63	68.8
Blueprint-Bench 2	6.7	26.5
browsecomp	76.2	85.9
charxiv_reasoning_with_tools	85.3	83.2
charxiv_rq	72.4	89.9
coding_arena_elo	1523	1447
critpt	3.1	17.7
DeepSWE 1.1	30	12
Finance Agent v2	51	43
finance_agent	63.3	59.7
frontiermath_tier_4	8.3	16.7
gdpval	54.8	23.3
GDPval-AA (Elo)	1676	1317
GDPval-AA v2	1381	962
GPQA Diamond	89.9	94.3
HLE	49	51.4
HLE (with tools)	46.8	51.6
Humanity’s Last Exam	33.2	44.4
IFBench	56.6	77.1
itbenchSre	39.8	30.3
Legal Agent Benchmark	5.4	0
livebench	75.5	79.9
livebench_agentic_coding	42.6	44.1
livebench_coding	79.3	76.5
livebench_data_analysis	78	78.5
livebench_instruction_following	63.2	79.1
livebench_language	76.1	85.4
livebench_math	87	91
livebench_reasoning	84.8	84
MCP Atlas	69.5	78.2
mmmlu	89.3	92.6
MMMU-Pro	75.6	83
mrcr_v2_8needle_128k_average	84.9	84.9
officeqa_pro	53.4	42.9
OmniScience Accuracy	40.9	55.3
OmniScience Non-Hallucination	51.6	50.1
OSWorld-Verified	78.5	76.2
scicode	47	59
simpleqa_verified	29	77.3
SWE-bench Pro	58.1	54.2
SWE-bench Verified	79.6	80.6
TauBench V3 - Banking	34.4	21.4
Terminal-Bench 2.0	59.1	68.5
Terminal-Bench 2.1	71.2	74
Terminal-Bench Hard	59.1	68.5
toolathlon	49.4	48.8
usamo_2026	55	74.4
vectara_answer_rate	99.9	99.4
vectara_avg_summary_length	114.7	107.7
vectara_factual_consistency	89.4	89.6
vectara_hallucination_rate ↓	10.6	10.4
τ²-Bench Telecom (AA run)	97.9	99.3
τ³-Bench	30.5	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.