Which is better, Gemini 3 Flash Preview or GPT-5?

Across 45 shared benchmarks tracked on Vector Wire, Gemini 3 Flash Preview leads on 34 and GPT-5 leads on 10 — Gemini 3 Flash Preview leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemini 3 Flash Preview vs GPT-5

GooglevsOpenAI45 shared benchmarks34–10 head-to-head

Benchmark	Gemini 3 Flash Preview	GPT-5
AA Agentic Index	49.7	49.7
AA Intelligence	38.7	35.3
AA-LCR	73	76.3
AA-Omniscience	10.1	-8.7
AIME 2025	99.7	95
arc_agi_1	84.7	56.2
ARC-AGI-2	33.6	7.5
arena_vision	1260	1212
Artificial Analysis Coding Index	42.6	38.9
coding_arena_elo	1387	1419
critpt	8.6	5.7
frontiermath_tier_4	4.2	12.5
gaia_level1	82.8	78.5
gaia_level2	66	64.8
gaia_level3	49	46.9
gaia_overall	68.4	66.1
gdpval	35.2	32.5
GPQA Diamond	90.4	87.3
harmbench	70	70
HLE	43.5	28.5
hmmt_feb_2025	97.5	88.3
hmmt_nov_2025	93.3	89.2
IFBench	78	73.1
matharena_visual_math_overall	85.8	78.8
MMMU-Pro	81.2	78.4
OmniScience Accuracy	53.4	40.3
OmniScience Non-Hallucination	7.6	21
scicode	50.6	43
simplebench	61.1	56.7
simpleqa_verified	67.4	50.6
swe_bench_bash	75.8	65
SWE-bench Multilingual	72.7	55.3
SWE-bench Pro	49.6	41.8
SWE-bench Verified	78	74.9
TauBench V3 - Banking	20.8	22.1
Terminal-Bench 2.0	47.6	35.2
Terminal-Bench 2.1	58	35.2
Terminal-Bench Hard	38.6	37.9
vectara_answer_rate	99.8	99.9
vectara_avg_summary_length	90.2	162.7
vectara_factual_consistency	86.5	85.3
vectara_hallucination_rate ↓	13.5	14.7
VideoMMMU	86.9	84.6
τ²-Bench Telecom (AA run)	80.4	86.5
τ³-Bench	17.5	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.