Which is better, Gemini 2.5 Pro or GPT-4o?

Across 50 shared benchmarks tracked on Vector Wire, Gemini 2.5 Pro leads on 44 and GPT-4o leads on 6 — Gemini 2.5 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 GPT-4o

GooglevsOpenAI50 shared benchmarks44–6 head-to-head

Benchmark	Gemini 2.5 Pro	GPT-4o
AA Agentic Index	7.2	9.7
AA Intelligence	27	11.2
AA-LCR	66	55
AA-Omniscience	-14.3	-10.5
aider_polyglot	82.2	30.7
AIME 2025	88.3	11.7
AIR-Bench 2024	73.6	62.4
anthropic_red_team	99.5	99.1
arc_agi_1	41	4.5
ARC-AGI-2	4.9	0
arena_vision	1246	1162
Artificial Analysis Coding Index	46.7	24.2
bbq	96.4	95.1
critpt	2.6	0
Fortress	54.9	47.2
gdpval	8.5	0
GPQA Diamond	84.4	70.1
harmbench	65.4	82.9
HLE	22.5	5.3
IFBench	49	36
LiveCodeBench	82.7	38.3
livecodebench_easy	98.8	82.5
livecodebench_hard	59.4	4.5
livecodebench_medium	90.6	32.1
mathvision	73.3	30.4
MathVista	83.9	63.8
MMLU-Pro	86	74.7
MMMU	83.9	72.2
MMMU-Pro	74.9	59.9
MMStar	79.2	63.9
multichallenge	53.6	40.3
OCRBench	85.9	806
OmniScience Accuracy	39	23.7
OmniScience Non-Hallucination	12.6	62.1
PropensityBench	79	46.1
scicode	43	33.4
SEAL VISTA	50.8	34.9
simple_safety_tests	97	98.5
simplebench	62.4	17.8
simpleqa	50.8	39.4
swe_bench_bash	53.6	21.6
SWE-bench Verified	67.2	38.8
Terminal-Bench Hard	26.5	8.3
vectara_answer_rate	99.1	93.8
vectara_avg_summary_length	106.4	86.6
vectara_factual_consistency	93	90.4
vectara_hallucination_rate ↓	7	9.6
Video-MME	84.8	77.2
xstest	98.7	97.3
τ²-Bench Telecom (AA run)	54.1	28.9

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.