Which is better, Claude Sonnet 4 or Gemini 2.5 Flash?

Across 45 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 24 and Gemini 2.5 Flash leads on 21 — Claude Sonnet 4 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 vs Gemini 2.5 Flash

AnthropicvsGoogle45 shared benchmarks24–21 head-to-head

Benchmark	Claude Sonnet 4	Gemini 2.5 Flash
AA Agentic Index	39.2	18.8
AA Intelligence	29.8	17.7
AA-LCR	69.7	65.7
AA-Omniscience	0.2	-29.8
aider_polyglot	70.7	61.9
AIME	33.1	56.3
AIME 2024	43.4	82.3
AIME 2025	74	72
AIR-Bench 2024	88.3	68.7
anthropic_red_team	99.2	98.8
arc_agi_1	63.7	33.3
ARC-AGI-2	13.6	2.5
arena_vision	1207	1214
Artificial Analysis Coding Index	37.6	22.2
bbq	97.9	97.7
critpt	1.1	1.4
frontiermath_tier_4	0	4.2
gdpval	31.2	11.9
GPQA Diamond	78	82.8
harmbench	98	62.6
HLE	10.7	12.1
IFBench	55	50.3
LiveCodeBench	68.5	76.2
livecodebench_easy	98.8	99.1
livecodebench_hard	32.6	48.3
livecodebench_medium	76	82.5
MMMU	74.4	79.7
MMMU-Pro	62.4	69.1
OmniScience Accuracy	22.7	26.1
OmniScience Non-Hallucination	70.9	24.7
scicode	40	35.9
SEAL VISTA	45.5	49.1
simple_safety_tests	100	98
simplebench	45.5	41.2
simpleqa	15.9	26.9
swe_bench_bash	64.9	28.7
SWE-bench Verified	80.2	60.4
Terminal-Bench Hard	31.1	13.6
vectara_answer_rate	98.6	99
vectara_avg_summary_length	145.8	101.5
vectara_factual_consistency	89.7	92.2
vectara_hallucination_rate ↓	10.3	7.8
VisualToolBench	4.5	4.7
xstest	97.2	98.8
τ²-Bench Telecom (AA run)	64.6	31.6

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.