Which is better, Kimi K2.6 or Qwen3.5 35B A3B?

Across 52 shared benchmarks tracked on Vector Wire, Kimi K2.6 leads on 47 and Qwen3.5 35B A3B leads on 5 — 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 35B A3B

MoonshotvsAlibaba52 shared benchmarks47–5 head-to-head

Benchmark	Kimi K2.6	Qwen3.5 35B A3B
AA Agentic Index	58.7	44.1
AA Intelligence	45.1	29.9
AA-LCR	76.7	68.3
AA-Omniscience	6.4	-48.1
AgentWorldBench - Android	58.9	53.2
AgentWorldBench - MCP	65.2	57.9
AgentWorldBench - OS	60.8	56.3
AgentWorldBench - Overall	53.4	47.7
AgentWorldBench - Search	27.5	26
AgentWorldBench - SWE	58.8	47.6
AgentWorldBench - Terminal	52.5	46.1
AgentWorldBench - Web	50.2	47.1
aime_2026	96.4	93.3
Artificial Analysis Coding Index	61.8	37
baby_vision	68.5	38.4
browsecomp	86.3	61
charxiv_rq	86.7	77.5
Claw Eval (pass@3)	80.9	51
coding_arena_elo	1509	1250
critpt	8	0.9
gdpval	41.4	20.3
GPQA Diamond	91.1	84.5
HLE	37.5	47.4
hmmt_feb_2026	94.7	81.8
IFBench	76	72.5
imo_answer_bench	86	76.8
itbenchSre	31.2	21.5
LiveCodeBench v6	89.6	74.6
mathvision	93.2	83.9
MCP Atlas	68.1	62.4
mcpmark	55.9	27
MMLU-Pro	87.1	85.3
MMMU-Pro	80.1	75.1
OJBench	60.6	36
OmniScience Accuracy	32.8	20.1
OmniScience Non-Hallucination	60.7	14.6
OSWorld-Verified	73.1	54.5
scicode	53.5	37.7
SWE-bench Multilingual	76.7	60.3
SWE-bench Pro	58.6	44.6
SWE-bench Verified	80.2	70
TauBench V3 - Banking	23.3	4.9
Terminal-Bench 2.0	66.7	40.5
Terminal-Bench 2.1	65.9	40.8
Terminal-Bench Hard	43.9	26.5
vectara_answer_rate	99.7	99.8
vectara_avg_summary_length	116.7	94.9
vectara_factual_consistency	89.2	89.5
vectara_hallucination_rate ↓	10.8	10.5
WideSearch	80.8	59.1
τ²-Bench Telecom (AA run)	95.9	89.2
τ³-Bench	20.6	68.9

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.