Which is better, Gemini 3.1 Pro or GPT-5.2?

Across 66 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 53 and GPT-5.2 leads on 12 — Gemini 3.1 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 3.1 Pro vs GPT-5.2

GooglevsOpenAI66 shared benchmarks53–12 head-to-head

Benchmark	Gemini 3.1 Pro	GPT-5.2
AA Agentic Index	23	60.2
AA Intelligence	47.7	43.3
AA-LCR	79	79.3
AA-Omniscience	32.9	0.3
aime_2026	98.3	98.3
apexAgents	33.5	23
arc_agi_1	98	86.2
ARC-AGI-2	77.1	52.9
arena_elo	1486	1437
arena_text_factuality	1471	1442
arena_vision	1280	1245
Artificial Analysis Coding Index	68.8	48.7
browsecomp	85.9	65.8
browsecomp_with_context_manager	85.9	65.8
charxiv_rq	89.9	82.1
coding_arena_elo	1447	1418
critpt	17.7	11.6
deepsearchqa_f1	81.9	71.3
Fortress	29.8	17.5
frontiermath_tier_4	16.7	18.8
gdpval	23.3	48.3
GDPval-AA (Elo)	1317	1462
GPQA Diamond	94.3	92.4
HLE	51.4	45.5
HLE (with tools)	51.6	45.5
hmmt_feb_2026	94.7	97
hmmt_nov_2025	94.8	99.2
IFBench	77.1	75.4
imo_answer_bench	91	86.3
livebench	79.9	74.8
livebench_agentic_coding	44.1	50.3
livebench_coding	76.5	76.1
livebench_data_analysis	78.5	78.2
livebench_instruction_following	79.1	61.8
livebench_language	85.4	79.8
livebench_math	91	93.2
livebench_reasoning	84	83.2
LiveCodeBench	91.7	89
LiveCodeBench Pro (Elo)	2887	2393
matharena_visual_math_overall	89.4	86.5
mathvision	89.8	83
MCP Atlas	78.2	68
MMLU-Pro	91	87
mmmlu	92.6	89.6
MMMU-Pro	83	79.5
mrcr_v2_8needle_128k_average	84.9	83.8
MultiNRC	64.7	42.2
OmniScience Accuracy	55.3	44.3
OmniScience Non-Hallucination	50.1	38.4
scicode	59	52.1
simplebench	79.6	45.8
simpleqa_verified	77.3	38.9
SWE-bench Multilingual	76.9	72
SWE-bench Pro	54.2	55.6
SWE-bench Verified	80.6	80
TauBench V3 - Banking	21.4	11.1
Terminal-Bench 2.0	68.5	54
Terminal-Bench Hard	68.5	62.2
tool_decathlon	48.8	46.3
toolathlon	48.8	46.3
vectara_answer_rate	99.4	100
vectara_avg_summary_length	107.7	186.3
vectara_factual_consistency	89.6	89.2
vectara_hallucination_rate ↓	10.4	10.8
τ²-Bench Telecom (AA run)	99.3	98.7
τ³-Bench	67.1	12

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.