Which is better, Gemini 3 Pro or GPT-5.2?

Across 88 shared benchmarks tracked on Vector Wire, Gemini 3 Pro leads on 40 and GPT-5.2 leads on 46 — GPT-5.2 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 Pro vs GPT-5.2

GooglevsOpenAI88 shared benchmarks40–46 head-to-head

Benchmark	Gemini 3 Pro	GPT-5.2
AA Agentic Index	52	60.2
AA Intelligence	40.6	43.3
AA-LCR	73	79.3
AA-Omniscience	15.3	0.3
AIME 2025	100	100
aime_2026	91.7	98.3
AIME25 no tools	96	98
apexAgents	18.4	23
arc_agi_1	75	86.2
ARC-AGI-2	54	52.9
arena_elo	1485	1437
arena_text_factuality	1481	1442
arena_vision	1290	1245
Artificial Analysis Coding Index	46.5	48.7
browsecomp	59.2	65.8
BrowseComp (context management)	67.6	70
browsecomp_with_context_manager	59.2	65.8
browsecomp_zh	66.8	76.1
charxiv_rq	81.4	82.1
coding_arena_elo	1438	1418
critpt	9.1	11.6
deepsearchqa_f1	63.2	71.3
Fortress	41.7	17.5
frontiermath_tier_4	18.8	18.8
gdpval	34.2	48.3
GDPval-AA (Elo)	1195	1462
GPQA Diamond	91.9	92.4
HLE	45.8	45.5
HLE (with tools)	45.8	45.5
hmmt_feb_2025	97.5	100
hmmt_feb_2026	86.4	97
hmmt_nov_2025	93.3	99.2
IFBench	70.4	75.4
imo_answer_bench	83.3	86.3
InfoVQA (val)	57.2	84
livebench	73.4	74.8
LiveCodeBench	92	89
LiveCodeBench Pro (Elo)	2439	2393
longbench_v2	68.2	54.5
LongVideoBench	77.7	76.5
matharena_visual_math_overall	84.2	86.5
mathvision	86.1	83
MathVista	89.8	82.8
MCP Atlas	70.3	68
metr_hcast	71.3	77
metr_re_bench	35	35
metr_swaa	99.5	99.6
MMLU-Pro	90.1	87
mmmlu	91.8	89.6
MMMU-Pro	81	79.5
MMVU	78.9	80.8
MotionBench	70.3	64.8
mrcr_v2_8needle_128k_average	77	83.8
MultiNRC	59	42.2
OCRBench	90.3	80.7
ocrbench_v2	63.4	50.5
OmniDocBench 1.5	88.5	85.7
OmniScience Accuracy	55.8	44.3
OmniScience Non-Hallucination	10	38.4
PropensityBench	52.9	34.4
scicode	56.1	52.1
screenspot_pro_no_tools	72.7	86.3
SEAL VISTA	51.5	46.6
Seal-0	45.5	45
simplebench	76.4	45.8
simpleqa_verified	72.9	38.9
SimpleVQA	69.7	55.8
swe_bench_bash	74.2	72.8
SWE-bench Multilingual	68.7	72
SWE-bench Pro	43.3	55.6
SWE-bench Verified	78	80
SWE-Perf	6.5	3.6
SWT-bench	79.7	80.7
Terminal-Bench 2.0	54.2	54
Terminal-Bench Hard	56.9	62.2
tool_decathlon	36.4	46.3
toolathlon	36.4	46.3
vectara_answer_rate	99.4	100
vectara_avg_summary_length	101.9	186.3
vectara_factual_consistency	86.4	89.2
vectara_hallucination_rate ↓	13.6	10.8
Video-MME	88.4	86
VideoMMMU	87.6	85.9
WorldVQA	47.4	28
ZeroBench	8	9
ZeroBench (w/ tools)	12	7
τ²-Bench	98	85.5
τ²-Bench Telecom (AA run)	98	98.7

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.