Which is better, DeepSeek-V4-Pro or Kimi K2.6?

Across 72 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Pro leads on 47 and Kimi K2.6 leads on 25 — DeepSeek-V4-Pro leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

DeepSeek-V4-Pro vs Kimi K2.6

DeepSeekvsMoonshot72 shared benchmarks47–25 head-to-head

Benchmark	DeepSeek-V4-Pro	Kimi K2.6
AA Agentic Index	63.3	58.7
AA Intelligence	53	45.1
AA-LCR	70	76.7
AA-Omniscience	-10.6	6.4
AgentWorldBench - Android	55.2	58.9
AgentWorldBench - MCP	63.3	65.2
AgentWorldBench - OS	63.7	60.8
AgentWorldBench - Overall	53	53.4
AgentWorldBench - Search	27.6	27.5
AgentWorldBench - SWE	59.4	58.8
AgentWorldBench - Terminal	51.3	52.5
AgentWorldBench - Web	50.3	50.2
aime_2026	96.7	96.4
Apex (Pass@1)	38.3	24
Apex Shortlist (Pass@1)	90.2	75.5
apexAgents	24.3	28.5
arena_elo	1457	1461
arena_text_factuality	1450	1455
Artificial Analysis Coding Index	59.4	61.8
browsecomp	83.4	86.3
BrowseComp (w/ Ctx)	83.4	83.2
Chinese SimpleQA (C-SimpleQA)	77.7	75.9
coding_arena_elo	1582	1509
critpt	13	8
FORTRESS (Adversarial)	36	65.6
FORTRESS (Benign)	98.5	97.2
gdpval	49	41.4
GDPval-AA (Elo)	1554	1482
GDPval-AA v2	1307	1190
Global-MMLU-Lite	89.3	88.4
GPQA Diamond	90.5	91.1
HLE	48.2	37.5
HLE (with tools)	48.2	54
hmmt_feb_2026	95.2	94.7
IFBench	76.5	76
imo_answer_bench	89.8	86
itbenchSre	38.3	31.2
livebench	73.6	72.2
livebench_agentic_coding	42.6	46.9
livebench_coding	70	78.6
livebench_data_analysis	74.5	65.1
livebench_instruction_following	62.4	64.4
livebench_language	78.1	75.1
livebench_math	90.7	84.3
livebench_reasoning	82.7	79.4
LiveCodeBench	93.5	89.6
MCP Atlas	74.2	68.1
MCPAtlas Public (Pass@1)	73.6	66.6
MMLU-Pro	87.5	87.1
OmniScience Accuracy	43	32.8
OmniScience Non-Hallucination	12.2	60.7
scicode	50	53.5
simpleqa_verified	57.9	38.7
strongreject	98.6	99.8
SWE-bench Multilingual	76.2	76.7
SWE-bench Pro	55.4	58.6
SWE-bench Verified	80.6	80.2
SWEBench Pro (Public)	55.4	58.6
Tau 3 Banking	26	20.6
TauBench V3 - Banking	30.1	23.3
Terminal Bench 2.1 (Best Harness)	64	71.3
Terminal-Bench 2.0	67.9	66.7
Terminal-Bench 2.1	72.1	65.9
Terminal-Bench Hard	46.2	43.9
toolathlon	51.8	50
usamo_2026	60.7	51.2
vectara_answer_rate	97.2	99.7
vectara_avg_summary_length	153.8	116.7
vectara_factual_consistency	91.4	89.2
vectara_hallucination_rate ↓	8.6	10.8
τ²-Bench Telecom (AA run)	96.2	95.9
τ³-Bench	25.8	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.