Which is better, Claude Opus 4.7 or Kimi K2.5?

Across 46 shared benchmarks tracked on Vector Wire, Claude Opus 4.7 leads on 42 and Kimi K2.5 leads on 3 — Claude Opus 4.7 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 4.7 vs Kimi K2.5

AnthropicvsMoonshot46 shared benchmarks42–3 head-to-head

Benchmark	Claude Opus 4.7	Kimi K2.5
AA Agentic Index	64.6	52.8
AA Intelligence	55	36
AA-LCR	75.3	73
AA-Omniscience	27.3	-7.3
aime_2026	95.8	95.8
arc_agi_1	92	65.3
ARC-AGI-2	75.8	11.8
arena_elo	1494	1450
arena_text_factuality	1474	1445
arena_vision	1304	1247
Artificial Analysis Coding Index	73.6	46.8
browsecomp	79.8	74.9
charxiv_rq	82.1	78.7
coding_arena_elo	1557	1436
critpt	12	3.1
cybergym	73.8	41.3
frontiermath_tier_4	22.9	4.2
gdpval	58.6	38.3
GPQA Diamond	94.2	87.9
HLE	54.7	50.2
HLE (with tools)	54.7	51.8
hmmt_feb_2026	93.9	87.1
IFBench	58.6	70.2
livebench	76.9	69.1
MCP Atlas	79.1	64
MMMU-Pro	78.8	78.5
OmniScience Accuracy	48.9	35.2
OmniScience Non-Hallucination	57.7	50
OSWorld-Verified	82.8	63.3
scicode	54.5	49
simplebench	61.7	46.8
simpleqa_verified	50.6	36.9
SWE-bench Multilingual	80.5	73
SWE-bench Pro	64.3	53.8
SWE-bench Verified	87.6	76.8
TauBench V3 - Banking	34.6	14.2
Terminal-Bench 2.0	69.4	50.8
Terminal-Bench 2.1	83.1	45.7
Terminal-Bench Hard	54.5	34.8
toolathlon	59.3	27.8
vectara_answer_rate	98	92.2
vectara_avg_summary_length	149.1	112
vectara_factual_consistency	88	85.8
vectara_hallucination_rate ↓	12	14.2
τ²-Bench Telecom (AA run)	88.6	95.9
τ³-Bench	28.9	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.