Which is better, Claude Fable 5 or Kimi K2.5?

Across 51 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 46 and Kimi K2.5 leads on 5 — Claude Fable 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 Fable 5 vs Kimi K2.5

AnthropicvsMoonshot51 shared benchmarks46–5 head-to-head

Benchmark	Claude Fable 5	Kimi K2.5
AA Agentic Index	56.6	52.8
AA Intelligence	62.1	36
AA-LCR	76.7	73
AA-Omniscience	43.3	-7.3
arc_agi_1	90.5	65.3
ARC-AGI-2	76.8	11.8
arena_elo	1508	1450
arena_text_factuality	1486	1445
arena_vision	1307	1247
Artificial Analysis Coding Index	76.5	46.8
baby_vision_with_python	90.5	40.5
browsecomp	88	74.9
BrowseComp (w/ Ctx)	88	74.9
charxiv_rq	89.4	78.7
coding_arena_elo	1626	1436
critpt	28.6	3.1
deepsearchqa_f1	94.2	89
FORTRESS (Adversarial)	96	54.1
FORTRESS (Benign)	55.1	98.3
gdpval	61.9	38.3
GDPval-AA v2	1760	1009
Global-MMLU-Lite	93.3	84
GPQA Diamond	92.6	87.9
HLE	64.5	50.2
HLE (with tools)	64.5	51.8
IFBench	63.5	70.2
livebench	78.3	69.1
mathvision	94.8	84.2
MCP Atlas	84.7	64
MMMU-Pro	84.2	78.5
OmniScience Accuracy	65.3	35.2
OmniScience Non-Hallucination	36.4	50
OSWorld-Verified	85	63.3
PaperBench	88.8	63.5
scicode	60.2	49
simplebench	81.9	46.8
simpleqa_verified	68.3	36.9
strongreject	98.7	99.5
SWE-bench Multilingual	86.6	73
SWE-bench Pro	80	53.8
SWE-bench Verified	95	76.8
SWEBench Pro (Public)	80	50.7
Tau 3 Banking	26.8	14.2
TauBench V3 - Banking	38.1	14.2
Terminal Bench 2.1 (Best Harness)	84.6	51.3
Terminal-Bench 2.1	88	45.7
Terminal-Bench Hard	62.9	34.8
toolathlon	61.7	27.8
WideSearch	81.2	79
τ²-Bench Telecom (AA run)	98.5	95.9
τ³-Bench	26.8	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.