Which is better, Claude Sonnet 4 or Kimi K2.5?

Across 47 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 11 and Kimi K2.5 leads on 36 — 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 —

Claude Sonnet 4 vs Kimi K2.5

AnthropicvsMoonshot47 shared benchmarks11–36 head-to-head

Benchmark	Claude Sonnet 4	Kimi K2.5
AA Agentic Index	39.2	52.8
AA Intelligence	29.8	36
AA-LCR	69.7	73
AA-Omniscience	0.2	-7.3
AIME 2025	74	96.1
arc_agi_1	63.7	65.3
ARC-AGI-2	13.6	11.8
arena_vision	1207	1247
Artificial Analysis Coding Index	37.6	46.8
browsecomp	12.2	74.9
browsecomp_zh	29.1	62.3
critpt	1.1	3.1
Fortress	24.4	41.1
frontiermath_tier_4	0	4.2
gdpval	31.2	38.3
GPQA Diamond	78	87.9
HLE	10.7	50.2
HLE (with tools)	20.3	51.8
HMMT 2025	15.9	95.4
IFBench	55	70.2
livebench	74.8	69.1
LiveCodeBench v6	48.5	85
MMLU-Pro	84	87.1
MMMU-Pro	62.4	78.5
multichallenge	57.1	61.4
MultiNRC	18.4	35.2
OmniScience Accuracy	22.7	35.2
OmniScience Non-Hallucination	70.9	50
OSWorld-Verified	42.2	63.3
scicode	40	49
SEAL VISTA	45.5	41.9
simplebench	45.5	46.8
swe_bench_bash	64.9	70.8
SWE-bench Multilingual	56.9	73
SWE-bench Pro	42.7	53.8
SWE-bench Verified	80.2	76.8
TauBench V3 - Banking	16.7	14.2
Terminal-Bench 2.1	36.3	45.7
Terminal-Bench Hard	31.1	34.8
vectara_answer_rate	98.6	92.2
vectara_avg_summary_length	145.8	112
vectara_factual_consistency	89.7	85.8
vectara_hallucination_rate ↓	10.3	14.2
xbench_deepsearch	64.6	76.7
τ²-Bench	65	85.4
τ²-Bench Telecom (AA run)	64.6	95.9
τ³-Bench	13.8	66

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.