Which is better, Kimi K2.5 or Kimi K2 (Reasoning)?

Across 44 shared benchmarks tracked on Vector Wire, Kimi K2.5 leads on 41 and Kimi K2 (Reasoning) leads on 2 — 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 Kimi K2 (Reasoning)

MoonshotvsMoonshot44 shared benchmarks41–2 head-to-head

Benchmark	Kimi K2.5	Kimi K2 (Reasoning)
AA Agentic Index	52.8	47.9
AA Intelligence	36	33.5
AA-LCR	73	70.3
AA-Omniscience	-7.3	-21.4
AIME 2025	96.1	94.5
Artificial Analysis Coding Index	46.8	34.8
browsecomp	74.9	60.2
browsecomp_with_context_manager	74.9	60.2
browsecomp_zh	62.3	62.3
critpt	3.1	2.6
frontiermath_tier_4	4.2	0
gaia_no_file	75.9	75.6
gdpval	38.3	24.5
GPQA Diamond	87.9	84.5
HLE	50.2	23.9
HLE (with tools)	51.8	44.9
HMMT 2025	95.4	89.4
hmmt_feb_2025	95.4	93.3
hmmt_nov_2025	91.1	89.2
IFBench	70.2	68.1
imo_answer_bench	81.8	78.6
LiveCodeBench v6	85	83.1
longbench_v2	61	45.1
MMLU-Pro	87.1	84.6
multichallenge	61.4	66.4
OJ-Bench (cpp)	57.4	48.7
OmniScience Accuracy	35.2	30.9
OmniScience Non-Hallucination	50	25.8
ResearchRubrics	59.5	56.2
scicode	49	44.8
Seal-0	57.4	56.3
simplebench	46.8	39.6
swe_bench_bash	70.8	63.4
SWE-bench Multilingual	73	61.1
SWE-bench Verified	76.8	71.3
Terminal-Bench 2.0	50.8	35.7
Terminal-Bench Hard	34.8	31.1
vectara_answer_rate	92.2	98.6
vectara_avg_summary_length	112	59.2
vectara_factual_consistency	85.8	82.1
vectara_hallucination_rate ↓	14.2	17.9
xbench_deepsearch	76.7	76
τ²-Bench	85.4	74.3
τ²-Bench Telecom (AA run)	95.9	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.