Which is better, DeepSeek-V3.2 or Kimi K2.6?

Across 44 shared benchmarks tracked on Vector Wire, DeepSeek-V3.2 leads on 5 and Kimi K2.6 leads on 39 — Kimi K2.6 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V3.2 vs Kimi K2.6

DeepSeekvsMoonshot44 shared benchmarks5–39 head-to-head

Benchmark	DeepSeek-V3.2	Kimi K2.6
AA Agentic Index	39.8	58.7
AA Intelligence	32.8	45.1
AA-LCR	70.7	76.7
AA-Omniscience	-22.5	6.4
aime_2026	95.1	96.4
apexAgents	14.5	28.5
arena_text_factuality	1433	1455
Artificial Analysis Coding Index	44.2	61.8
browsecomp	51.4	86.3
coding_arena_elo	1368	1509
critpt	2.9	8
deepsearchqa_f1	60.9	92.5
frontiermath_tier_4	2.1	14.6
gdpval	18.8	41.4
GPQA Diamond	84	91.1
HLE	40.8	37.5
HLE (with tools)	40.8	54
hmmt_feb_2026	84.1	94.7
IFBench	61	76
imo_answer_bench	78.3	86
LiveCodeBench	86	89.6
LiveCodeBench v6	83.3	89.6
MCP Atlas	62.2	68.1
mcpmark	38	55.9
MMLU-Pro	86	87.1
OmniScience Accuracy	33	32.8
OmniScience Non-Hallucination	17.3	60.7
scicode	39	53.5
simpleqa_verified	27.5	38.7
SWE-bench Multilingual	70.2	76.7
SWE-bench Pro	15.6	58.6
SWE-bench Verified	73.1	80.2
TauBench V3 - Banking	18.8	23.3
Terminal-Bench 2.0	46.4	66.7
Terminal-Bench 2.1	46.8	65.9
Terminal-Bench Hard	35.6	43.9
toolathlon	35.2	50
vectara_answer_rate	92.6	99.7
vectara_avg_summary_length	62	116.7
vectara_factual_consistency	93.7	89.2
vectara_hallucination_rate ↓	6.3	10.8
WideSearch (item-f1)	32.5	80.8
τ²-Bench Telecom (AA run)	90.6	95.9
τ³-Bench	69.2	20.6

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.