Which is better, Gemini 3.1 Pro or Kimi K3?

Across 54 shared benchmarks tracked on Vector Wire, Gemini 3.1 Pro leads on 13 and Kimi K3 leads on 41 — Kimi K3 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 Kimi K3

GooglevsMoonshot54 shared benchmarks13–41 head-to-head

Benchmark	Gemini 3.1 Pro	Kimi K3
AA Agentic Index	23	54.3
AA Intelligence	47.7	60
AA-LCR	79	82.7
AA-Omniscience	32.9	46
aime_2026	98.3	96.7
apex_agents	33.5	41
apexAgents	33.5	41.3
arc_agi_1	98	65.7
ARC-AGI-2	77.1	12.4
arena_elo	1486	1489
arena_text_factuality	1471	1471
Artificial Analysis Coding Index	68.8	76.2
baby_vision	51.6	85.7
baby_vision_with_python	68.3	85.7
browsecomp	85.9	91.2
charxiv_rq	89.9	84.8
coding_arena_elo	1447	1674
critpt	17.7	23.4
cybergym	38.8	80
deepsearchqa_f1	81.9	95
DeepSWE	10	67.5
DeepSWE 1.1	12	69
Finance Agent v2	43	54.4
gdpval	23.3	59.4
GDPval-AA v2	962	1687
GPQA Diamond	94.3	93.5
HLE	51.4	56
hmmt_feb_2026	94.7	97
itbenchSre	30.3	47.7
livebench_agentic_coding	44.1	62.2
livebench_coding	76.5	81.5
livebench_data_analysis	78.5	78.7
livebench_instruction_following	79.1	71.4
livebench_language	85.4	85.5
livebench_math	91	84.4
livebench_reasoning	84	90.7
matharena_visual_math_overall	89.4	91.3
mathvision	89.8	97.8
MCP Atlas	78.2	84.2
MMMU-Pro	83	83.4
officeqa_pro	42.9	63.3
OmniScience Accuracy	55.3	47.6
OmniScience Non-Hallucination	50.1	49.1
OSWorld-Verified	76.2	84.8
PostTrainBench	21.6	36.6
Program Bench	39.5	77.8
scicode	59	58.7
simplebench	79.6	60.7
simpleqa_verified	77.3	42.7
SWE-Marathon	4	42
Tau 3 Banking	16.5	33.4
TauBench V3 - Banking	21.4	46
Terminal-Bench 2.1	74	88.3
toolathlon	48.8	73.2

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.