Which is better, Gemini 3.1 Pro or Gemini 3.6 Flash?

Across 42 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 21 and Gemini 3.6 Flash leads on 20 — 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 3.6 Flash

GooglevsGoogle42 shared benchmarks21–20 head-to-head

Benchmark	Gemini 3.1 Pro	Gemini 3.6 Flash
AA Agentic Index	23	40.5
AA Intelligence	47.7	52
AA-LCR	79	79
AA-Omniscience	32.9	22.1
aime_2026	98.3	96.7
arc_agi_1	98	34.5
ARC-AGI-2	77.1	2.6
arena_elo	1486	1481
arena_text_factuality	1471	1470
Artificial Analysis Coding Index	68.8	69.2
charxiv_reasoning_with_tools	83.2	89.4
charxiv_rq	89.9	85.2
coding_arena_elo	1447	1538
critpt	17.7	10.6
DeepSWE	10	49
DeepSWE 1.1	12	49
GDM-MRCR v2 (8-needle) (1M (pointwise))	26.3	54
gdpval	23.3	46.1
GDPval-AA v2	962	1421
GDPVal-AA v2 (Elo)	965	1422
GPQA Diamond	94.3	92.8
HLE	51.4	40.8
hmmt_feb_2026	94.7	89.4
livebench_agentic_coding	44.1	43.4
livebench_coding	76.5	77.9
livebench_data_analysis	78.5	63
livebench_instruction_following	79.1	75.4
livebench_language	85.4	83.9
livebench_math	91	86.4
livebench_reasoning	84	85.2
matharena_visual_math_overall	89.4	87.8
MLE-Bench	42.6	63.9
MMMU-Pro	83	83.2
mrcr_v2_8needle_128k_average	84.9	91.8
OmniScience Accuracy	55.3	50
OmniScience Non-Hallucination	50.1	44.4
OSWorld-Verified	76.2	83
scicode	59	52.7
simpleqa_verified	77.3	68.7
SWE-bench Pro	54.2	58.7
TauBench V3 - Banking	21.4	29.9
Terminal-Bench 2.1	74	78

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.