Which is better, Claude Opus 4.5 or Kimi K2.5?

Across 81 shared benchmarks tracked on Vector Wire, Claude Opus 4.5 leads on 44 and Kimi K2.5 leads on 37 — Claude Opus 4.5 leads on more. Strengths differ by axis; the table below shows every head-to-head number with provenance.

SEEDRefresh Models —Deals —Regulatory —Sources —

Claude Opus 4.5 vs Kimi K2.5

AnthropicvsMoonshot81 shared benchmarks44–37 head-to-head

Benchmark	Claude Opus 4.5	Kimi K2.5
AA Agentic Index	59.6	52.8
AA Intelligence	41.9	36
AA-LCR	76	73
AA-Omniscience	14	-7.3
AIME 2025	92.8	96.1
AIME 2026 I	93.3	92.5
aime_2026	95.1	95.8
arc_agi_1	82	65.3
ARC-AGI-2	37.6	11.8
arena_elo	1473	1450
arena_text_factuality	1458	1445
Artificial Analysis Coding Index	47.8	46.8
browsecomp	37	74.9
BrowseComp (context management)	59.2	74.9
browsecomp_with_context_manager	67.8	74.9
browsecomp_zh	62.4	62.3
charxiv_rq	68.5	78.7
coding_arena_elo	1495	1436
critpt	4.6	3.1
cybergym	50.6	41.3
deepsearchqa_f1	76.1	89
FinSearchCompT2&T3	66.2	67.8
Fortress	13.6	41.1
frontiermath_tier_4	4.2	4.2
gdpval	47.3	38.3
GPQA Diamond	87	87.9
HLE	30.8	50.2
HLE (with tools)	43.4	51.8
hmmt_feb_2025	92.9	95.4
hmmt_feb_2026	85.3	87.1
hmmt_nov_2025	93.3	91.1
IFBench	58	70.2
imo_answer_bench	84	81.8
InfoVQA (val)	76.9	92.6
livebench	76	69.1
LiveCodeBench v6	84.8	85
longbench_v2	64.4	61
LongVideoBench	67.2	79.8
mathvision	77.1	84.2
MathVista	80.2	90.1
MCP Atlas	69.8	64
MMLU-Pro	90	87.1
MMMU-Pro	74	78.5
MMVU	77.3	80.4
MotionBench	60.3	70.4
multichallenge	59	61.4
MultiNRC	48.6	35.2
nl2repo	43.2	32
OCRBench	86.5	92.3
OJ-Bench (cpp)	54.6	57.4
OJBench (cpp)	54.6	57.4
OmniDocBench 1.5	87.7	88.8
OmniScience Accuracy	46.6	35.2
OmniScience Non-Hallucination	39	50
OSWorld-Verified	66.3	63.3
PaperBench	72.9	63.5
scicode	50	49
SEAL VISTA	46.4	41.9
Seal-0	47.7	57.4
simplebench	62	46.8
simpleqa_verified	41.8	36.9
SimpleVQA	69.7	71.2
swe_bench_bash	76.8	70.8
SWE-bench Multilingual	77.5	73
SWE-bench Pro	57.1	53.8
SWE-bench Verified	81.5	76.8
Terminal-Bench 2.0	59.3	50.8
Terminal-Bench Hard	47	34.8
tool_decathlon	43.5	27.8
toolathlon	43.5	27.8
vectara_answer_rate	98.7	92.2
vectara_avg_summary_length	114.5	112
vectara_factual_consistency	89.1	85.8
vectara_hallucination_rate ↓	10.9	14.2
VideoMMMU	84.4	86.6
WideSearch (item-f1)	76.2	72.7
WorldVQA	36.8	46.3
ZeroBench	3	9
ZeroBench (w/ tools)	9	11
τ²-Bench	98.2	85.4
τ²-Bench Telecom (AA run)	89.5	95.9

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.