Which is better, Gemini 2.5 Pro or Gemini 3 Pro?

Across 52 shared benchmarks tracked on Vector Wire, Gemini 2.5 Pro leads on 8 and Gemini 3 Pro leads on 44 — Gemini 3 Pro leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemini 2.5 Pro vs Gemini 3 Pro

GooglevsGoogle52 shared benchmarks8–44 head-to-head

Benchmark	Gemini 2.5 Pro	Gemini 3 Pro
AA Agentic Index	7.2	52
AA Intelligence	27	40.6
AA-LCR	66	73
AA-Omniscience	-14.3	15.3
AIME 2025	88.3	100
arc_agi_1	41	75
ARC-AGI-2	4.9	54
arena_elo	1446	1485
arena_text_factuality	1462	1481
arena_vision	1246	1290
Artificial Analysis Coding Index	46.7	46.5
browsecomp	9.9	59.2
browsecomp_zh	32.2	66.8
coding_arena_elo	1224	1438
critpt	2.6	9.1
Fortress	54.9	41.7
frontiermath_tier_4	4.2	18.8
gdpval	8.5	34.2
GPQA Diamond	84.4	91.9
HLE	22.5	45.8
HLE (with tools)	28.4	45.8
hmmt_feb_2025	82.5	97.5
hmmt_nov_2025	80	93.3
IFBench	49	70.4
LiveCodeBench	82.7	92
matharena_visual_math_overall	77.2	84.2
mathvision	73.3	86.1
MathVista	83.9	89.8
MMLU-Pro	86	90.1
MMMU-Pro	74.9	81
mrcr	93	89.7
multichallenge	53.6	65.7
OCRBench	85.9	90.3
ocrbench_v2	59.3	63.4
OmniScience Accuracy	39	55.8
OmniScience Non-Hallucination	12.6	10
PropensityBench	79	52.9
scicode	43	56.1
SEAL VISTA	50.8	51.5
simplebench	62.4	76.4
simpleqa	50.8	72.1
simpleqa_verified	56	72.9
swe_bench_bash	53.6	74.2
SWE-bench Verified	67.2	78
Terminal-Bench Hard	26.5	56.9
vectara_answer_rate	99.1	99.4
vectara_avg_summary_length	106.4	101.9
vectara_factual_consistency	93	86.4
vectara_hallucination_rate ↓	7	13.6
Video-MME	84.8	88.4
τ²-Bench	54	98
τ²-Bench Telecom (AA run)	54.1	98

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.