Which is better, Claude Sonnet 4 or Kimi K2 (Reasoning)?

Across 44 shared benchmarks tracked on Vector Wire, Claude Sonnet 4 leads on 16 and Kimi K2 (Reasoning) leads on 25 — Kimi K2 (Reasoning) 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 (Reasoning)

AnthropicvsMoonshot44 shared benchmarks16–25 head-to-head

Benchmark	Claude Sonnet 4	Kimi K2 (Reasoning)
AA Agentic Index	39.2	47.9
AA Intelligence	29.8	33.5
AA-LCR	69.7	70.3
AA-Omniscience	0.2	-21.4
aider_polyglot	70.7	59.1
AIME 2025	74	94.5
ArtifactsBench	57.3	54.2
Artificial Analysis Coding Index	37.6	34.8
browsecomp	12.2	60.2
browsecomp_zh	29.1	62.3
critpt	1.1	2.6
FinSearchComp-global	42	29.5
frontiermath_tier_4	0	0
GAIA (text only)	68.3	60.2
gdpval	31.2	24.5
GPQA Diamond	78	84.5
HLE	10.7	23.9
HLE (with tools)	20.3	44.9
HMMT 2025	15.9	89.4
IFBench	55	68.1
LiveCodeBench	68.5	79.2
LiveCodeBench v6	48.5	83.1
mmlu_redux	93.6	94.4
MMLU-Pro	84	84.6
Multi-SWE-Bench	35.7	41.9
multichallenge	57.1	66.4
OmniScience Accuracy	22.7	30.9
OmniScience Non-Hallucination	70.9	25.8
scicode	40	44.8
simplebench	45.5	39.6
swe_bench_bash	64.9	63.4
SWE-bench Multilingual	56.9	61.1
SWE-bench Verified	80.2	71.3
SWE-Dev	67.1	66.6
Terminal-Bench	36.4	47.1
Terminal-Bench Hard	31.1	31.1
theagentcompany	37	30
vectara_answer_rate	98.6	98.6
vectara_avg_summary_length	145.8	59.2
vectara_factual_consistency	89.7	82.1
vectara_hallucination_rate ↓	10.3	17.9
xbench_deepsearch	64.6	76
τ²-Bench	65	74.3
τ²-Bench Telecom (AA run)	64.6	93

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.