Which is better, Kimi K2.6 or Qwen3.5 397B A17B?

Across 55 shared benchmarks tracked on Vector Wire, Kimi K2.6 leads on 45 and Qwen3.5 397B A17B leads on 10 — 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 —

Kimi K2.6 vs Qwen3.5 397B A17B

MoonshotvsAlibaba55 shared benchmarks45–10 head-to-head

Benchmark	Kimi K2.6	Qwen3.5 397B A17B
AA Agentic Index	58.7	53.3
AA Intelligence	45.1	34.3
AA-LCR	76.7	72.7
AA-Omniscience	6.4	-30.8
AgentWorldBench - Android	58.9	54.9
AgentWorldBench - MCP	65.2	68.3
AgentWorldBench - OS	60.8	60.9
AgentWorldBench - Overall	53.4	54.7
AgentWorldBench - Search	27.5	30.8
AgentWorldBench - SWE	58.8	64.4
AgentWorldBench - Terminal	52.5	55.3
AgentWorldBench - Web	50.2	48.5
aime_2026	96.4	94.2
apexAgents	28.5	15.3
arena_elo	1461	1442
arena_text_factuality	1455	1445
arena_vision	1264	1249
Artificial Analysis Coding Index	61.8	48.2
browsecomp	86.3	69
charxiv_rq	86.7	80.8
Claw Eval (pass@3)	80.9	48.1
coding_arena_elo	1509	1399
critpt	8	1.7
gdpval	41.4	35.8
GDPval-AA v2	1190	962
Global-MMLU-Lite	88.4	90
GPQA Diamond	91.1	89.3
HLE	37.5	29
HLE (with tools)	54	48.3
hmmt_feb_2026	94.7	87.9
IFBench	76	78.8
imo_answer_bench	86	80.9
itbenchSre	31.2	34.1
LiveCodeBench v6	89.6	83.6
mcpmark	55.9	46.1
MMLU-Pro	87.1	87.8
MMMU-Pro	80.1	79
OmniScience Accuracy	32.8	30.8
OmniScience Non-Hallucination	60.7	17.3
scicode	53.5	42
simpleqa_verified	38.7	26
strongreject	99.8	99.4
SWE-bench Multilingual	76.7	69.3
SWE-bench Pro	58.6	50.9
SWE-bench Verified	80.2	76.4
Tau 3 Banking	20.6	13.4
TauBench V3 - Banking	23.3	13.4
Terminal-Bench 2.0	66.7	52.5
Terminal-Bench 2.1	65.9	51.3
Terminal-Bench Hard	43.9	40.9
toolathlon	50	38.3
usamo_2026	51.2	36.3
WideSearch	80.8	74
τ²-Bench Telecom (AA run)	95.9	95.6
τ³-Bench	20.6	13.4

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.