Which is better, Gemini 2.5 Pro or GPT-5?

Across 57 shared benchmarks tracked on Vector Wire, Gemini 2.5 Pro leads on 9 and GPT-5 leads on 47 — GPT-5 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-5

GooglevsOpenAI57 shared benchmarks9–47 head-to-head

Benchmark	Gemini 2.5 Pro	GPT-5
AA Agentic Index	7.2	49.7
AA Intelligence	27	35.3
AA-LCR	66	76.3
AA-Omniscience	-14.3	-8.7
aider_polyglot	82.2	88
AIME 2025	88.3	95
AIR-Bench 2024	73.6	87.7
arc_agi_1	41	56.2
ARC-AGI-2	4.9	7.5
arena_vision	1246	1212
ArtifactsBench	57.7	73
Artificial Analysis Coding Index	46.7	38.9
browsecomp	9.9	54.9
browsecomp_zh	32.2	65
coding_arena_elo	1224	1419
critpt	2.6	5.7
FinSearchComp-global	42.6	63.9
Fortress	54.9	17
frontiermath_tier_4	4.2	12.5
GAIA (text only)	60.2	76.4
gdpval	8.5	32.5
GPQA Diamond	84.4	87.3
harmbench	65.4	70
HLE	22.5	28.5
HLE (with tools)	28.4	35.2
HMMT 2025	65.7	93.3
hmmt_feb_2025	82.5	88.3
hmmt_nov_2025	80	89.2
IFBench	49	73.1
imo_2025	31.6	38.1
LiveCodeBench	82.7	85
matharena_visual_math_overall	77.2	78.8
MMLU-Pro	86	87.5
MMMU	83.9	84.2
MMMU-Pro	74.9	78.4
multichallenge	53.6	71.1
ocrbench_v2	59.3	55.5
OmniScience Accuracy	39	40.3
OmniScience Non-Hallucination	12.6	21
scicode	43	43
SEAL VISTA	50.8	49.7
simplebench	62.4	56.7
simpleqa_verified	56	50.6
swe_bench_bash	53.6	65
SWE-bench Verified	67.2	74.9
TauBench V3 - Banking	9.7	22.1
Terminal-Bench	25.3	43.8
Terminal-Bench 2.1	28.5	35.2
Terminal-Bench Hard	26.5	37.9
vectara_answer_rate	99.1	99.9
vectara_avg_summary_length	106.4	162.7
vectara_factual_consistency	93	85.3
vectara_hallucination_rate ↓	7	14.7
xbench_deepsearch	56	77.8
τ²-Bench	54	85
τ²-Bench Telecom (AA run)	54.1	86.5
τ³-Bench	9.3	19.6

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.