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

Across 58 shared benchmarks tracked on Vector Wire, Kimi K2.5 leads on 49 and Qwen3.5 35B A3B leads on 9 — Kimi K2.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Kimi K2.5 vs Qwen3.5 35B A3B

MoonshotvsAlibaba58 shared benchmarks49–9 head-to-head

Benchmark	Kimi K2.5	Qwen3.5 35B A3B
AA Agentic Index	52.8	44.1
AA Intelligence	36	29.9
AA-LCR	73	68.3
AA-Omniscience	-7.3	-48.1
aime_2026	95.8	93.3
Artificial Analysis Coding Index	46.8	37
baby_vision	36.5	38.4
browsecomp	74.9	61
browsecomp_zh	62.3	69.5
charxiv_rq	78.7	77.5
coding_arena_elo	1436	1250
critpt	3.1	0.9
gdpval	38.3	20.3
GPQA Diamond	87.9	84.5
HLE	50.2	47.4
HMMT 2025	95.4	89.2
hmmt_feb_2025	95.4	89
hmmt_feb_2026	87.1	81.8
hmmt_nov_2025	91.1	89.2
IFBench	70.2	72.5
imo_answer_bench	81.8	76.8
LiveCodeBench v6	85	74.6
longbench_v2	61	59
LVBench	75.9	71.4
mathvision	84.2	83.9
MathVista	90.1	86.2
MCP Atlas	64	62.4
mcpmark	29.5	27
MMLU-Pro	87.1	85.3
MMMU-Pro	78.5	75.1
MMVU	80.4	72.3
multichallenge	61.4	60
nl2repo	32	20.5
OCRBench	92.3	91
OmniDocBench 1.5	88.8	89.3
OmniScience Accuracy	35.2	20.1
OmniScience Non-Hallucination	50	14.6
OSWorld-Verified	63.3	54.5
scicode	49	37.7
Seal-0	57.4	41.4
SimpleVQA	71.2	58.3
SWE-bench Multilingual	73	60.3
SWE-bench Pro	53.8	44.6
SWE-bench Verified	76.8	70
TauBench V3 - Banking	14.2	4.9
Terminal-Bench 2.0	50.8	40.5
Terminal-Bench 2.1	45.7	40.8
Terminal-Bench Hard	34.8	26.5
tool_decathlon	27.8	28.7
vectara_answer_rate	92.2	99.8
vectara_avg_summary_length	112	94.9
vectara_factual_consistency	85.8	89.5
vectara_hallucination_rate ↓	14.2	10.5
VideoMMMU	86.6	80.4
WideSearch	79	59.1
ZeroBench	9	0.1
τ²-Bench Telecom (AA run)	95.9	89.2
τ³-Bench	66	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.