Which is better, Claude Opus 4.6 or Kimi K2.5?

Across 72 shared benchmarks tracked on Vector Wire, Claude Opus 4.6 leads on 57 and Kimi K2.5 leads on 15 — Claude Opus 4.6 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 4.6 vs Kimi K2.5

AnthropicvsMoonshot72 shared benchmarks57–15 head-to-head

Benchmark	Claude Opus 4.6	Kimi K2.5
AA Agentic Index	67.6	52.8
AA Intelligence	44.9	36
AA-LCR	74.3	73
AA-Omniscience	13.7	-7.3
AIME 2025	99.8	96.1
aime_2026	96.7	95.8
apexAgents	33	11.5
arc_agi_1	93	65.3
ARC-AGI-2	68.8	11.8
arena_elo	1497	1450
arena_text_factuality	1487	1445
arena_vision	1300	1247
Artificial Analysis Coding Index	48.1	46.8
baby_vision	14.8	36.5
baby_vision_with_python	38.4	40.5
browsecomp	86.8	74.9
BrowseComp (Agent Swarm)	83.7	78.4
browsecomp_with_context_manager	84	74.9
charxiv_rq	69.1	78.7
charxiv_rq_with_python	84.7	78.7
coding_arena_elo	1545	1436
critpt	12.6	3.1
cybergym	73.8	41.3
deepsearchqa_accuracy	80.6	77.1
deepsearchqa_f1	91.3	89
Fortress	20.5	41.1
frontiermath_tier_4	22.9	4.2
gdpval	55.9	38.3
GPQA Diamond	91.3	87.9
HLE	53.1	50.2
HLE (with tools)	62.7	51.8
hmmt_feb_2026	96.2	87.1
hmmt_nov_2025	96.3	91.1
IFBench	62.5	70.2
imo_answer_bench	75.3	81.8
livebench	76.3	69.1
LiveCodeBench v6	88.8	85
matharena_visual_math_overall	72.3	80.6
mathvision	71.2	84.2
mathvision_with_python	84.6	85
MCP Atlas	76.8	64
mcpmark	56.7	29.5
MMLU-Pro	89.1	87.1
mmmu_pro_with_python	77.3	77.7
MMMU-Pro	77.3	78.5
multichallenge	56	61.4
MultiNRC	57.1	35.2
nl2repo	49.8	32
OJBench (python)	60.3	54.7
OmniDocBench 1.5	86.6	88.8
OmniScience Accuracy	47	35.2
OmniScience Non-Hallucination	37.2	50
OSWorld-Verified	72.7	63.3
scicode	52	49
SEAL VISTA	46.1	41.9
simplebench	67.6	46.8
simpleqa_verified	46.5	36.9
swe_bench_bash	75.6	70.8
SWE-bench Multilingual	77.8	73
SWE-bench Pro	57.3	53.8
SWE-bench Verified	80.8	76.8
Terminal-Bench 2.0	65.4	50.8
Terminal-Bench Hard	65.4	34.8
tool_decathlon	47.2	27.8
toolathlon	56.8	27.8
V* (w/ python)	86.4	86.9
vectara_answer_rate	99.8	92.2
vectara_avg_summary_length	137.6	112
vectara_factual_consistency	87.8	85.8
vectara_hallucination_rate ↓	12.2	14.2
τ²-Bench Telecom (AA run)	99.3	95.9
τ³-Bench	72.4	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.