Which is better, Gemini 3 Pro or Kimi K2 (Reasoning)?

Across 46 shared benchmarks tracked on Vector Wire, Gemini 3 Pro leads on 41 and Kimi K2 (Reasoning) leads on 5 — Gemini 3 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 Pro vs Kimi K2 (Reasoning)

GooglevsMoonshot46 shared benchmarks41–5 head-to-head

Benchmark	Gemini 3 Pro	Kimi K2 (Reasoning)
AA Agentic Index	52	47.9
AA Intelligence	40.6	33.5
AA-LCR	73	70.3
AA-Omniscience	15.3	-21.4
AIME 2025	100	94.5
AIME25 no tools	96	94.5
Arena-Hard (Creative Writing)	93.6	80.1
Arena-Hard (Hard Prompt)	72.6	71.9
Artificial Analysis Coding Index	46.5	34.8
browsecomp	59.2	60.2
browsecomp_with_context_manager	59.2	60.2
browsecomp_zh	66.8	62.3
critpt	9.1	2.6
frontiermath_tier_4	18.8	0
gdpval	34.2	24.5
GPQA Diamond	91.9	84.5
HLE	45.8	23.9
HLE (with tools)	45.8	44.9
hmmt_feb_2025	97.5	93.3
hmmt_nov_2025	93.3	89.2
IFBench	70.4	68.1
imo_answer_bench	83.3	78.6
LiveCodeBench	92	79.2
LiveCodeBench v6	90.7	83.1
longbench_v2	68.2	45.1
MMLU-Pro	90.1	84.6
mrcr	89.7	44.2
Multi-SWE-Bench	42.7	41.9
multichallenge	65.7	66.4
OJ-Bench (cpp)	68.5	48.7
OmniScience Accuracy	55.8	30.9
OmniScience Non-Hallucination	10	25.8
scicode	56.1	44.8
Seal-0	45.5	56.3
simplebench	76.4	39.6
swe_bench_bash	74.2	63.4
SWE-bench Multilingual	68.7	61.1
SWE-bench Verified	78	71.3
Terminal-Bench 2.0	54.2	35.7
Terminal-Bench Hard	56.9	31.1
vectara_answer_rate	99.4	98.6
vectara_avg_summary_length	101.9	59.2
vectara_factual_consistency	86.4	82.1
vectara_hallucination_rate ↓	13.6	17.9
τ²-Bench	98	74.3
τ²-Bench Telecom (AA run)	98	93

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.