Which is better, Gemini 3.1 Pro or GPT-5?

Across 46 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 41 and GPT-5 leads on 5 — 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

GooglevsOpenAI46 shared benchmarks41–5 head-to-head

Benchmark	Gemini 3.1 Pro	GPT-5
AA Agentic Index	23	49.7
AA Intelligence	47.7	35.3
AA-LCR	79	76.3
AA-Omniscience	32.9	-8.7
arc_agi_1	98	56.2
ARC-AGI-2	77.1	7.5
arena_vision	1280	1212
Artificial Analysis Coding Index	68.8	38.9
browsecomp	85.9	54.9
coding_arena_elo	1447	1419
critpt	17.7	5.7
Fortress	29.8	17
frontiermath_tier_4	16.7	12.5
gdpval	23.3	32.5
GPQA Diamond	94.3	87.3
HLE	51.4	28.5
HLE (with tools)	51.6	35.2
hmmt_nov_2025	94.8	89.2
IFBench	77.1	73.1
imo_answer_bench	91	76
LiveCodeBench	91.7	85
LiveCodeBench v6	91.7	87
matharena_visual_math_overall	89.4	78.8
MMLU-Pro	91	87.5
MMMU-Pro	83	78.4
multichallenge	71.4	71.1
MultiNRC	64.7	52.1
OmniScience Accuracy	55.3	40.3
OmniScience Non-Hallucination	50.1	21
scicode	59	43
simplebench	79.6	56.7
simpleqa_verified	77.3	50.6
SWE-bench Multilingual	76.9	55.3
SWE-bench Pro	54.2	41.8
SWE-bench Verified	80.6	74.9
TauBench V3 - Banking	21.4	22.1
Terminal-Bench 2.0	68.5	35.2
Terminal-Bench 2.1	74	35.2
Terminal-Bench Hard	68.5	37.9
vectara_answer_rate	99.4	99.9
vectara_avg_summary_length	107.7	162.7
vectara_factual_consistency	89.6	85.3
vectara_hallucination_rate ↓	10.4	14.7
VisualToolBench	29	18.7
τ²-Bench Telecom (AA run)	99.3	86.5
τ³-Bench	67.1	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.