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

Across 51 shared benchmarks tracked on Vector Wire, DeepSeek-V4-Flash leads on 27 and Kimi K2.6 leads on 23 — DeepSeek-V4-Flash 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-Flash vs Kimi K2.6

DeepSeekvsMoonshot51 shared benchmarks27–23 head-to-head

Benchmark	DeepSeek-V4-Flash	Kimi K2.6
AA Agentic Index	61.3	58.7
AA Intelligence	51.8	45.1
AA-LCR	74.3	76.7
AA-Omniscience	-14.3	6.4
aime_2026	95.8	96.4
Apex (Pass@1)	33	24
Apex Shortlist (Pass@1)	85.7	75.5
arena_elo	1439	1461
arena_text_factuality	1434	1455
Artificial Analysis Coding Index	69.1	61.8
browsecomp	73.2	86.3
Chinese SimpleQA (C-SimpleQA)	78.9	75.9
coding_arena_elo	1431	1509
critpt	16.6	8
gdpval	52.9	41.4
GDPval-AA (Elo)	1395	1482
GDPval-AA v2	1189	1190
Global-MMLU-Lite	88.4	88.4
GPQA Diamond	90.8	91.1
HLE	38.6	37.5
HLE (with tools)	45.1	54
hmmt_feb_2026	94.8	94.7
IFBench	79.2	76
imo_answer_bench	88.4	86
itbenchSre	31.5	31.2
livebench_agentic_coding	46.8	46.9
livebench_coding	75	78.6
livebench_data_analysis	79.3	65.1
livebench_instruction_following	65.5	64.4
livebench_language	79.2	75.1
livebench_math	86.8	84.3
livebench_reasoning	86.6	79.4
LiveCodeBench	91.6	89.6
MCP Atlas	69	68.1
MMLU-Pro	86.4	87.1
OmniScience Accuracy	40.4	32.8
OmniScience Non-Hallucination	15.6	60.7
scicode	49.9	53.5
simpleqa_verified	45	38.7
strongreject	97.4	99.8
SWE-bench Multilingual	73.3	76.7
SWE-bench Pro	52.6	58.6
SWE-bench Verified	79	80.2
Tau 3 Banking	22.9	20.6
TauBench V3 - Banking	39.4	23.3
Terminal-Bench 2.0	56.9	66.7
Terminal-Bench 2.1	82.7	65.9
Terminal-Bench Hard	38.6	43.9
toolathlon	70.3	50
τ²-Bench Telecom (AA run)	95.6	95.9
τ³-Bench	22.9	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.