Which is better, DeepSeek-V3.2 or Gemini 2.5 Pro?

Across 47 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 31 and Gemini 2.5 Pro leads on 15 — DeepSeek-V3.2 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V3.2 vs Gemini 2.5 Pro

DeepSeekvsGoogle47 shared benchmarks31–15 head-to-head

Benchmark	DeepSeek-V3.2	Gemini 2.5 Pro
AA Agentic Index	39.8	7.2
AA Intelligence	32.8	27
AA-LCR	70.7	66
AA-Omniscience	-22.5	-14.3
AIME 2025	93.1	88.3
arc_agi_1	57	41
ARC-AGI-2	4	4.9
arena_text_factuality	1433	1462
ArtifactsBench	55.8	57.7
Artificial Analysis Coding Index	44.2	46.7
browsecomp	51.4	9.9
browsecomp_zh	65	32.2
coding_arena_elo	1368	1224
critpt	2.9	2.6
FinSearchComp-global	26.2	42.6
frontiermath_tier_4	2.1	4.2
GAIA (text only)	63.5	60.2
gdpval	18.8	8.5
GPQA Diamond	84	84.4
HLE	40.8	22.5
HLE (with tools)	40.8	28.4
HMMT 2025	90.2	65.7
hmmt_feb_2025	92.5	82.5
hmmt_nov_2025	90.2	80
IFBench	61	49
LiveCodeBench	86	82.7
MMLU-Pro	86	86
mrcr	55.5	93
OmniScience Accuracy	33	39
OmniScience Non-Hallucination	17.3	12.6
scicode	39	43
simpleqa_verified	27.5	56
swe_bench_bash	60	53.6
SWE-bench Verified	73.1	67.2
TauBench V3 - Banking	18.8	9.7
Terminal-Bench	37.7	25.3
Terminal-Bench 2.1	46.8	28.5
Terminal-Bench Hard	35.6	26.5
theagentcompany	34	39.3
vectara_answer_rate	92.6	99.1
vectara_avg_summary_length	62	106.4
vectara_factual_consistency	93.7	93
vectara_hallucination_rate ↓	6.3	7
xbench_deepsearch	71	56
τ²-Bench	91	54
τ²-Bench Telecom (AA run)	90.6	54.1
τ³-Bench	69.2	9.3

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.