Which is better, Claude Sonnet 4.6 or Gemini 3 Flash Preview?

Across 48 shared benchmarks tracked on Vector Wire, Claude Sonnet 4.6 leads on 35 and Gemini 3 Flash Preview leads on 12 — Claude Sonnet 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 Sonnet 4.6 vs Gemini 3 Flash Preview

AnthropicvsGoogle48 shared benchmarks35–12 head-to-head

Benchmark	Claude Sonnet 4.6	Gemini 3 Flash Preview
AA Agentic Index	61.6	49.7
AA Intelligence	48.4	38.7
AA-LCR	74	73
AA-Omniscience	12.2	10.1
apexAgents	28	27.7
arc_agi_1	86.5	84.7
ARC-AGI-2	60.4	33.6
arena_elo	1472	1473
arena_text_factuality	1460	1452
arena_vision	1278	1260
Artificial Analysis Coding Index	63	42.6
Blueprint-Bench 2	6.7	0
charxiv_rq	72.4	80.3
coding_arena_elo	1523	1387
critpt	3.1	8.6
Finance Agent v2	51	42.6
frontiermath_tier_4	8.3	4.2
gdpval	54.8	35.2
GDPval-AA (Elo)	1676	1204
GPQA Diamond	89.9	90.4
harmbench	24.2	70
HLE	49	43.5
Humanity’s Last Exam	33.2	33.7
IFBench	56.6	78
Legal Agent Benchmark	5.4	0
livebench	75.5	72.4
MCP Atlas	69.5	62
mmmlu	89.3	91.8
MMMU-Pro	75.6	81.2
mrcr_v2_8needle_128k_average	84.9	67.2
OmniScience Accuracy	40.9	53.4
OmniScience Non-Hallucination	51.6	7.6
OSWorld-Verified	78.5	65.1
scicode	47	50.6
simpleqa_verified	29	67.4
SWE-bench Pro	58.1	49.6
SWE-bench Verified	79.6	78
TauBench V3 - Banking	34.4	20.8
Terminal-Bench 2.0	59.1	47.6
Terminal-Bench 2.1	71.2	58
Terminal-Bench Hard	59.1	38.6
toolathlon	49.4	49.4
vectara_answer_rate	99.9	99.8
vectara_avg_summary_length	114.7	90.2
vectara_factual_consistency	89.4	86.5
vectara_hallucination_rate ↓	10.6	13.5
τ²-Bench Telecom (AA run)	97.9	80.4
τ³-Bench	30.5	17.5

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.