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

Across 55 shared benchmarks tracked on Vector Wire, Kimi K2.5 leads on 32 and Qwen3.5 397B A17B leads on 23 — 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 397B A17B

MoonshotvsAlibaba55 shared benchmarks32–23 head-to-head

Benchmark	Kimi K2.5	Qwen3.5 397B A17B
AA Agentic Index	52.8	53.3
AA Intelligence	36	34.3
AA-LCR	73	72.7
AA-Omniscience	-7.3	-30.8
aime_2026	95.8	94.2
apexAgents	11.5	15.3
arena_elo	1450	1442
arena_text_factuality	1445	1445
arena_vision	1247	1249
Artificial Analysis Coding Index	46.8	48.2
browsecomp	74.9	69
BrowseComp (context management)	74.9	78.6
browsecomp_zh	62.3	70.3
charxiv_rq	78.7	80.8
coding_arena_elo	1436	1399
critpt	3.1	1.7
gdpval	38.3	35.8
GDPval-AA v2	1009	962
Global-MMLU-Lite	84	90
GPQA Diamond	87.9	89.3
HLE	50.2	29
HLE (with tools)	51.8	48.3
HMMT 2025	95.4	92.7
hmmt_feb_2025	95.4	94.8
hmmt_feb_2026	87.1	87.9
hmmt_nov_2025	91.1	92.7
IFBench	70.2	78.8
imo_answer_bench	81.8	80.9
LiveCodeBench v6	85	83.6
longbench_v2	61	63.2
mcpmark	29.5	46.1
MMLU-Pro	87.1	87.8
MMMU-Pro	78.5	79
multichallenge	61.4	67.6
nl2repo	32	32.2
OmniScience Accuracy	35.2	30.8
OmniScience Non-Hallucination	50	17.3
scicode	49	42
Seal-0	57.4	46.9
simpleqa_verified	36.9	26
SimpleVQA	71.2	67.1
strongreject	99.5	99.4
SWE-bench Multilingual	73	69.3
SWE-bench Pro	53.8	50.9
SWE-bench Verified	76.8	76.4
Tau 3 Banking	14.2	13.4
TauBench V3 - Banking	14.2	13.4
Terminal-Bench 2.0	50.8	52.5
Terminal-Bench 2.1	45.7	51.3
Terminal-Bench Hard	34.8	40.9
toolathlon	27.8	38.3
VideoMMMU	86.6	84.7
WideSearch	79	74
τ²-Bench Telecom (AA run)	95.9	95.6
τ³-Bench	66	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.