Which is better, Gemini 3.5 Flash or Gemini 3 Flash Preview?

Across 45 shared benchmarks tracked on Vector Wire, Gemini 3.5 Flash leads on 40 and Gemini 3 Flash Preview leads on 5 — Gemini 3.5 Flash leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Gemini 3.5 Flash vs Gemini 3 Flash Preview

GooglevsGoogle45 shared benchmarks40–5 head-to-head

Benchmark	Gemini 3.5 Flash	Gemini 3 Flash Preview
AA Agentic Index	70.4	49.7
AA Intelligence	52	38.7
AA-LCR	81	73
AA-Omniscience	21.2	10.1
aime_2026	95	96.7
apexAgents	47.1	27.7
arc_agi_1	92.5	84.7
ARC-AGI-2	72.1	33.6
arena_elo	1477	1473
arena_text_factuality	1466	1452
Artificial Analysis Coding Index	70.1	42.6
Blueprint-Bench 2	33.6	0
charxiv_rq	84.2	80.3
coding_arena_elo	1490	1387
critpt	13.1	8.6
Finance Agent v2	57.9	42.6
frontiermath_tier_4	14.6	4.2
gdpval	57.8	35.2
GDPval-AA (Elo)	1656	1204
GPQA Diamond	92.2	90.4
HLE	42.7	43.5
hmmt_feb_2026	95.5	89.4
Humanity’s Last Exam	40.2	33.7
IFBench	76.3	78
Legal Agent Benchmark	0.8	0
livebench	75	72.4
matharena_visual_math_overall	89.9	85.8
MCP Atlas	83.6	62
MMMU-Pro	84.3	81.2
mrcr_v2_8needle_128k_average	77.3	67.2
mrcr_v2_8needle_1m_pointwise	22.1	26.3
OmniScience Accuracy	51.4	53.4
OmniScience Non-Hallucination	38.2	7.6
OSWorld-Verified	78.4	65.1
scicode	53.1	50.6
simplebench	76.7	61.1
simpleqa_verified	68.4	67.4
SWE-bench Pro	55.1	49.6
TauBench V3 - Banking	32.2	20.8
Terminal-Bench 2.0	76.2	47.6
Terminal-Bench 2.1	78.7	58
Terminal-Bench Hard	46.2	38.6
toolathlon	56.5	49.4
τ²-Bench Telecom (AA run)	95.6	80.4
τ³-Bench	25.4	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.