Which is better, Gemini 3.1 Pro or Gemini 3.5 Flash?

Across 62 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 31 and Gemini 3.5 Flash leads on 31 — 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 —

Gemini 3.1 Pro vs Gemini 3.5 Flash

GooglevsGoogle62 shared benchmarks31–31 head-to-head

Benchmark	Gemini 3.1 Pro	Gemini 3.5 Flash
AA Agentic Index	23	70.4
AA Intelligence	47.7	52
AA-LCR	79	81
AA-Omniscience	32.9	21.2
aime_2026	98.3	95
apexAgents	33.5	47.1
arc_agi_1	98	92.5
ARC-AGI-2	77.1	72.1
arena_elo	1486	1477
arena_text_factuality	1471	1466
Artificial Analysis Coding Index	68.8	70.1
Blueprint-Bench 2	26.5	33.6
charxiv_reasoning_with_tools	83.2	84.9
charxiv_rq	89.9	84.2
coding_arena_elo	1447	1490
critpt	17.7	13.1
DeepSWE	10	37
DeepSWE 1.1	12	37
Finance Agent v2	43	57.9
finance_agent	59.7	57.9
frontiermath_tier_4	16.7	14.6
GDM-MRCR v2 (8-needle) (1M (pointwise))	26.3	26.6
gdpval	23.3	57.8
GDPval-AA (Elo)	1317	1656
GDPval-AA v2	962	1357
GDPVal-AA v2 (Elo)	965	1349
GPQA Diamond	94.3	92.2
HiL-Bench	35.3	27.7
HLE	51.4	42.7
hmmt_feb_2026	94.7	95.5
Humanity’s Last Exam	44.4	40.2
IFBench	77.1	76.3
itbenchSre	30.3	40.3
Legal Agent Benchmark	0	0.8
livebench	79.9	75
livebench_agentic_coding	44.1	49
livebench_coding	76.5	78.2
livebench_data_analysis	78.5	64.9
livebench_instruction_following	79.1	75.6
livebench_language	85.4	84.6
livebench_math	91	88.2
livebench_reasoning	84	82
matharena_visual_math_overall	89.4	89.9
MCP Atlas	78.2	83.6
MLE-Bench	42.6	49.7
MMMU-Pro	83	84.3
mrcr_v2_8needle_128k_average	84.9	77.3
mrcr_v2_8needle_1m_pointwise	26.3	22.1
OmniScience Accuracy	55.3	51.4
OmniScience Non-Hallucination	50.1	38.2
OSWorld-Verified	76.2	78.4
scicode	59	53.1
simplebench	79.6	76.7
simpleqa_verified	77.3	68.4
SWE-bench Pro	54.2	55.1
TauBench V3 - Banking	21.4	32.2
Terminal-Bench 2.0	68.5	76.2
Terminal-Bench 2.1	74	78.7
Terminal-Bench Hard	68.5	46.2
toolathlon	48.8	56.5
τ²-Bench Telecom (AA run)	99.3	95.6
τ³-Bench	67.1	25.4

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.