Which is better, DeepSeek-V3.2 or Gemini 3.1 Pro?

Across 49 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 4 and Gemini 3.1 Pro leads on 45 — 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 —

DeepSeek-V3.2 vs Gemini 3.1 Pro

DeepSeekvsGoogle49 shared benchmarks4–45 head-to-head

Benchmark	DeepSeek-V3.2	Gemini 3.1 Pro
AA Agentic Index	39.8	23
AA Intelligence	32.8	47.7
AA-LCR	70.7	79
AA-Omniscience	-22.5	32.9
aime_2026	95.1	98.3
apexAgents	14.5	33.5
arc_agi_1	57	98
ARC-AGI-2	4	77.1
arena_text_factuality	1433	1471
Artificial Analysis Coding Index	44.2	68.8
browsecomp	51.4	85.9
browsecomp_with_context_manager	67.6	85.9
coding_arena_elo	1368	1447
critpt	2.9	17.7
cybergym	17.3	38.8
deepsearchqa_f1	60.9	81.9
frontiermath_tier_4	2.1	16.7
gdpval	18.8	23.3
GPQA Diamond	84	94.3
HLE	40.8	51.4
HLE (with tools)	40.8	51.6
hmmt_feb_2026	84.1	94.7
hmmt_nov_2025	90.2	94.8
IFBench	61	77.1
imo_answer_bench	78.3	91
LiveCodeBench	86	91.7
LiveCodeBench v6	83.3	91.7
MCP Atlas	62.2	78.2
mcpmark	38	55.9
MMLU-Pro	86	91
OmniScience Accuracy	33	55.3
OmniScience Non-Hallucination	17.3	50.1
scicode	39	59
simpleqa_verified	27.5	77.3
SWE-bench Multilingual	70.2	76.9
SWE-bench Pro	15.6	54.2
SWE-bench Verified	73.1	80.6
TauBench V3 - Banking	18.8	21.4
Terminal-Bench 2.0	46.4	68.5
Terminal-Bench 2.1	46.8	74
Terminal-Bench Hard	35.6	68.5
tool_decathlon	35.2	48.8
toolathlon	35.2	48.8
vectara_answer_rate	92.6	99.4
vectara_avg_summary_length	62	107.7
vectara_factual_consistency	93.7	89.6
vectara_hallucination_rate ↓	6.3	10.4
τ²-Bench Telecom (AA run)	90.6	99.3
τ³-Bench	69.2	67.1

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.