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

Across 45 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 11 and Gemini 3 Flash Preview leads on 34 — Gemini 3 Flash Preview 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 Flash Preview

DeepSeekvsGoogle45 shared benchmarks11–34 head-to-head

Benchmark	DeepSeek-V3.2	Gemini 3 Flash Preview
AA Agentic Index	39.8	49.7
AA Intelligence	32.8	38.7
AA-LCR	70.7	73
AA-Omniscience	-22.5	10.1
AIME 2025	93.1	99.7
aime_2026	95.1	96.7
apexAgents	14.5	27.7
arc_agi_1	57	84.7
ARC-AGI-2	4	33.6
arena_text_factuality	1433	1452
Artificial Analysis Coding Index	44.2	42.6
coding_arena_elo	1368	1387
critpt	2.9	8.6
frontiermath_tier_4	2.1	4.2
gaia_level1	90.3	82.8
gaia_level2	73.6	66
gaia_level3	61.2	49
gaia_overall	76.7	68.4
gdpval	18.8	35.2
GPQA Diamond	84	90.4
HLE	40.8	43.5
hmmt_feb_2025	92.5	97.5
hmmt_feb_2026	84.1	89.4
hmmt_nov_2025	90.2	93.3
IFBench	61	78
MCP Atlas	62.2	62
OmniScience Accuracy	33	53.4
OmniScience Non-Hallucination	17.3	7.6
scicode	39	50.6
simpleqa_verified	27.5	67.4
swe_bench_bash	60	75.8
SWE-bench Multilingual	70.2	72.7
SWE-bench Pro	15.6	49.6
SWE-bench Verified	73.1	78
TauBench V3 - Banking	18.8	20.8
Terminal-Bench 2.0	46.4	47.6
Terminal-Bench 2.1	46.8	58
Terminal-Bench Hard	35.6	38.6
toolathlon	35.2	49.4
vectara_answer_rate	92.6	99.8
vectara_avg_summary_length	62	90.2
vectara_factual_consistency	93.7	86.5
vectara_hallucination_rate ↓	6.3	13.5
τ²-Bench Telecom (AA run)	90.6	80.4
τ³-Bench	69.2	17.5

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.