Which is better, DeepSeek-V3.2 or Gemini 3 Pro?

Across 65 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 10 and Gemini 3 Pro leads on 54 — Gemini 3 Pro 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 3 Pro

DeepSeekvsGoogle65 shared benchmarks10–54 head-to-head

Benchmark	DeepSeek-V3.2	Gemini 3 Pro
AA Agentic Index	39.8	52
AA Intelligence	32.8	40.6
AA-LCR	70.7	73
AA-Omniscience	-22.5	15.3
AIME 2025	93.1	100
AIME 2026 I	92.7	90.6
aime_2026	95.1	91.7
AIME25 no tools	89.3	96
apexAgents	14.5	18.4
arc_agi_1	57	75
ARC-AGI-2	4	54
arena_text_factuality	1433	1481
Arena-Hard (Creative Writing)	88.8	93.6
Arena-Hard (Hard Prompt)	53.4	72.6
Artificial Analysis Coding Index	44.2	46.5
browsecomp	51.4	59.2
BrowseComp (context management)	67.6	67.6
browsecomp_with_context_manager	67.6	59.2
browsecomp_zh	65	66.8
coding_arena_elo	1368	1438
critpt	2.9	9.1
cybergym	17.3	39.9
deepsearchqa_f1	60.9	63.2
FinSearchCompT2&T3	59.1	49.9
frontiermath_tier_4	2.1	18.8
gdpval	18.8	34.2
GPQA Diamond	84	91.9
HLE	40.8	45.8
HLE (with tools)	40.8	45.8
hmmt_feb_2025	92.5	97.5
hmmt_feb_2026	84.1	86.4
hmmt_nov_2025	90.2	93.3
IFBench	61	70.4
imo_answer_bench	78.3	83.3
LiveCodeBench	86	92
LiveCodeBench v6	83.3	90.7
longbench_v2	59.8	68.2
MCP Atlas	62.2	70.3
MMLU-Pro	86	90.1
mrcr	55.5	89.7
Multi-SWE-Bench	37.4	42.7
OctoCodingbench	26	22.9
OJ-Bench (cpp)	54.7	68.5
OmniScience Accuracy	33	55.8
OmniScience Non-Hallucination	17.3	10
scicode	39	56.1
Seal-0	49.5	45.5
simpleqa_verified	27.5	72.9
swe_bench_bash	60	74.2
SWE-bench Multilingual	70.2	68.7
SWE-bench Pro	15.6	43.3
SWE-bench Verified	73.1	78
SWE-Perf	0.9	6.5
SWT-bench	62	79.7
Terminal-Bench 2.0	46.4	54.2
Terminal-Bench Hard	35.6	56.9
tool_decathlon	35.2	36.4
toolathlon	35.2	36.4
vectara_answer_rate	92.6	99.4
vectara_avg_summary_length	62	101.9
vectara_factual_consistency	93.7	86.4
vectara_hallucination_rate ↓	6.3	13.6
WideSearch (item-f1)	32.5	57
τ²-Bench	91	98
τ²-Bench Telecom (AA run)	90.6	98

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.