Which is better, Kimi K2.5 or Qwen3.5 27B?

Across 57 shared benchmarks tracked on Vector Wire, Kimi K2.5 leads on 44 and Qwen3.5 27B leads on 13 — 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 27B

MoonshotvsAlibaba57 shared benchmarks44–13 head-to-head

Benchmark	Kimi K2.5	Qwen3.5 27B
AA Agentic Index	52.8	54.6
AA Intelligence	36	34.6
AA-LCR	73	72.3
AA-Omniscience	-7.3	-44
aime_2026	95.8	92.6
arena_vision	1247	1221
Artificial Analysis Coding Index	46.8	34.9
baby_vision	36.5	44.6
browsecomp	74.9	61
browsecomp_zh	62.3	62.1
charxiv_rq	78.7	79.5
coding_arena_elo	1436	1357
critpt	3.1	0.9
gdpval	38.3	33
GPQA Diamond	87.9	85.8
HLE	50.2	48.5
HMMT 2025	95.4	89.8
hmmt_feb_2025	95.4	92
hmmt_feb_2026	87.1	84.3
hmmt_nov_2025	91.1	89.8
IFBench	70.2	76.5
imo_answer_bench	81.8	79.9
LiveCodeBench v6	85	80.7
longbench_v2	61	60.6
LVBench	75.9	73.6
mathvision	84.2	86
MathVista	90.1	87.8
MCP Atlas	64	68.4
mcpmark	29.5	36.3
MMLU-Pro	87.1	86.1
MMMU-Pro	78.5	75
MMVU	80.4	73.3
multichallenge	61.4	60.8
nl2repo	32	27.3
OCRBench	92.3	89.4
OmniDocBench 1.5	88.8	88.9
OmniScience Accuracy	35.2	20.7
OmniScience Non-Hallucination	50	24.9
OSWorld-Verified	63.3	56.2
scicode	49	39.5
Seal-0	57.4	47.2
SimpleVQA	71.2	56
SWE-bench Multilingual	73	69.3
SWE-bench Pro	53.8	51.2
SWE-bench Verified	76.8	75
Terminal-Bench 2.0	50.8	41.6
Terminal-Bench Hard	34.8	32.6
tool_decathlon	27.8	31.5
vectara_answer_rate	92.2	99.8
vectara_avg_summary_length	112	94.4
vectara_factual_consistency	85.8	87.9
vectara_hallucination_rate ↓	14.2	12.1
VideoMMMU	86.6	82.3
WideSearch	79	66.4
ZeroBench	9	0.1
τ²-Bench Telecom (AA run)	95.9	93.9
τ³-Bench	66	68.4

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.