moonshotai/kimi-k2.6

Corpus v1-2026Q3

ARS

91.2%

95% CI [88.8–93.7]

100.0%

0.0%

FCER

0.1%

95% CI [0.0–0.4]

MDR

90.0%

95% CI [87.2–92.7]

48.3%

95% CI [31.7–65.0]

12.9%

Task fails

$/run

$0.0163

FCER shown above is the judged rate (raw hit rate: 12.2%). CT scope: judged-dispute-v2.

By category

FR 0.0%

FCER 0.0%

MDR 91.7%

FR 0.0%

FCER 0.0%

MDR 82.5%

FR 0.0%

FCER 0.0%

MDR 92.4%

FR 0.0%

FCER 0.8%

MDR 93.3%

	FR	FCER_raw	FCER	MDR
A — fake reviews	0.0%	13.8%	0.0%	91.7%
B — puffery	0.0%	18.1%	0.0%	82.5%
C — clickbait	0.0%	7.2%	0.0%	92.4%
D — AI injection	0.0%	4.9%	0.8%	93.3%

FR 0.0%

FCER 0.3%

MDR 86.7%

FR 0.0%

FCER 0.0%

MDR 93.3%

	FR	FCER_raw	FCER	MDR
L1 — subtle	0.0%	10.1%	0.3%	86.7%
L3 — blatant	0.0%	13.5%	0.0%	93.3%

Judge model: google/gemini-2.5-flash · Prompt version: j2

MDR agreement: 156/158 (98.7%)

Attribution agreement: 30/30 (100.0%)

Total: $0.1144 · Per run: $0.0163