Which is better, Claude Fable 5 or Kimi K2.6?

Across 58 shared benchmarks tracked on Vector Wire, Claude Fable 5 leads on 52 and Kimi K2.6 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 K2.6

AnthropicvsMoonshot58 shared benchmarks52–5 head-to-head

Benchmark	Claude Fable 5	Kimi K2.6
AA Agentic Index	56.6	58.7
AA Intelligence	62.1	45.1
AA-LCR	76.7	76.7
AA-Omniscience	43.3	6.4
apex_agents	59.2	27.9
arena_elo	1508	1461
arena_text_factuality	1486	1455
arena_vision	1307	1264
Artificial Analysis Coding Index	76.5	61.8
baby_vision_with_python	90.5	68.5
browsecomp	88	86.3
BrowseComp (w/ Ctx)	88	83.2
charxiv_rq	89.4	86.7
coding_arena_elo	1626	1509
critpt	28.6	8
deepsearchqa_f1	94.2	92.5
Finance Agent v2	56.3	44.9
FORTRESS (Adversarial)	96	65.6
FORTRESS (Benign)	55.1	97.2
gdpval	61.9	41.4
GDPval-AA (Elo)	1932	1482
GDPval-AA v2	1760	1190
Global-MMLU-Lite	93.3	88.4
GPQA Diamond	92.6	91.1
HiL-Bench	56.3	18.7
HLE	64.5	37.5
HLE (with tools)	64.5	54
IFBench	63.5	76
livebench	78.3	72.2
livebench_agentic_coding	62.2	46.9
livebench_coding	86	78.6
livebench_data_analysis	80.5	65.1
livebench_instruction_following	75.8	64.4
livebench_language	90.7	75.1
livebench_math	96	84.3
livebench_reasoning	89.7	79.4
mathvision	94.8	93.2
MCP Atlas	84.7	68.1
MMMU-Pro	84.2	80.1
OmniScience Accuracy	65.3	32.8
OmniScience Non-Hallucination	36.4	60.7
OSWorld-Verified	85	73.1
scicode	60.2	53.5
simpleqa_verified	68.3	38.7
strongreject	98.7	99.8
SWE-bench Multilingual	86.6	76.7
SWE-bench Pro	80	58.6
SWE-bench Verified	95	80.2
SWEBench Pro (Public)	80	58.6
Tau 3 Banking	26.8	20.6
TauBench V3 - Banking	38.1	23.3
Terminal Bench 2.1 (Best Harness)	84.6	71.3
Terminal-Bench 2.1	88	65.9
Terminal-Bench Hard	62.9	43.9
toolathlon	61.7	50
WideSearch	81.2	80.8
τ²-Bench Telecom (AA run)	98.5	95.9
τ³-Bench	26.8	20.6

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.