General

WildGuardTest (Decision Index) leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on WildGuardTest (Decision Index) is 95.4% by d1-ultra. 88 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	d1-ultra	95.4%	measured
2	GLiDE	95.1%	measured
3	Perplexity Decider v1.2 27B	94.4%	measured
4	mammoth-1	94.2%	estimated ± 1.4 pp, low confidence
5	Celeris-1 Decision	93.3%	measured
6	elephant-1	93.3%	measured
7	Perplexity Decider Flash v1.2 4B	93.3%	measured
8	Torchcast Decision 27B	93.2%	estimated ± 1.4 pp, medium confidence
9	Eikos-27B	93.2%	estimated ± 1.4 pp, medium confidence
10	Jev 1.14	93.1%	measured
11	Kev 27B	92.9%	estimated ± 1.4 pp, medium confidence
12	vLLM-SR Decision 2.0 Vega 27B	92.9%	estimated ± 1.4 pp, medium confidence
13	simple-jev · Qwen3.8-27B	92.5%	estimated ± 1.4 pp, medium confidence
14	deck31b (Gemma 4 31B)	92.5%	estimated ± 1.4 pp, medium confidence
15	Quyet-1.0-Large	92.2%	estimated ± 1.4 pp, medium confidence
16	JEV-27B	92.1%	estimated ± 1.4 pp, medium confidence
17	Bespoke Nimble 9B v3	92.0%	estimated ± 1.4 pp, medium confidence
18	JADE	92.0%	estimated ± 1.4 pp, medium confidence
19	Drex v2.0	91.7%	estimated ± 1.4 pp, medium confidence
20	Jevfire	91.6%	estimated ± 1.4 pp, medium confidence
21	Xor 1.2	91.5%	estimated ± 1.4 pp, medium confidence
22	Drex v1.5	91.4%	estimated ± 1.4 pp, medium confidence
23	decisio 0.4.0 (Gemma 4 12B)	91.2%	estimated ± 1.4 pp, medium confidence
24	Decider chat · Gemma-4-31B	91.2%	estimated ± 1.4 pp, medium confidence
25	Decider chat · Qwen3.6-27B	91.0%	estimated ± 1.4 pp, medium confidence
26	Clef-Flash	90.9%	measured
27	d1	90.9%	measured
28	Blink v0.3 26B-A4B (NVFP4)	90.9%	estimated ± 1.4 pp, medium confidence
29	Surogate Rune 26B-A4B v3	90.7%	estimated ± 1.4 pp, medium confidence
30	Clef	90.7%	measured
31	mmastrac diffusiongemma (vLLM PR 57250)	90.6%	estimated ± 1.7 pp, medium confidence
32	decisio 0.4.0 (Qwen3.6-35B-A3B)	90.1%	estimated ± 1.4 pp, medium confidence
33	JoshuaSP diffusiongemma (open-jev)	90.1%	estimated ± 1.4 pp, medium confidence
34	Hopper 12B (trained)	90.0%	estimated ± 1.4 pp, medium confidence
35	JPT-35B-A3B	90.0%	estimated ± 1.4 pp, medium confidence
36	GEV-26B-Decide	90.0%	estimated ± 1.4 pp, medium confidence
37	vLLM-SR Decision 2.0 Lux 9B	89.6%	estimated ± 1.4 pp, medium confidence
38	Lev1 3.7B	89.2%	estimated ± 1.4 pp, medium confidence
39	Jev-Omni	89.1%	estimated ± 1.4 pp, medium confidence
40	Winnow-12B Q8	88.9%	estimated ± 1.4 pp, medium confidence
41	ezjev 4B s2	88.8%	estimated ± 1.4 pp, medium confidence
42	Bosun v3.1 1.7B	88.7%	estimated ± 4.6 pp, low confidence
43	Hopper (G) 1.2	88.7%	estimated ± 1.4 pp, medium confidence
44	Sieve 9B	88.6%	estimated ± 1.4 pp, medium confidence
45	open-jev (pngwn)	88.4%	estimated ± 1.7 pp, low confidence
46	JPT-9B	88.3%	estimated ± 1.4 pp, medium confidence
47	JevK5 9B v0.3.3	88.2%	estimated ± 1.4 pp, medium confidence
48	Microsoft-Decision-1	88.2%	measured
49	Mercury Decide	88.0%	measured
50	Kodiak v0.2 1B (accuracy)	87.5%	estimated ± 4.6 pp, low confidence
51	Kev 9B v2	87.5%	estimated ± 1.4 pp, low confidence
52	vLLM-SR Decision 2.0 Nox 4B	87.4%	estimated ± 1.4 pp, low confidence
53	Bekko System One v0 68M	87.3%	estimated ± 4.6 pp, low confidence
54	Bosun v3.1 0.6B	87.3%	estimated ± 4.6 pp, low confidence
55	lev	87.2%	estimated ± 1.4 pp, low confidence
56	Clef-Omni	87.2%	measured
57	vLLM-SR Decision 2.0 Kai 0.6B	87.1%	estimated ± 4.6 pp, low confidence
58	Jevstral 8B	87.1%	estimated ± 1.4 pp, low confidence
59	Sifr 0.8B v3.1	87.0%	estimated ± 4.6 pp, low confidence
60	JEV-9B	86.8%	estimated ± 1.4 pp, low confidence
61	mini-jev	86.7%	estimated ± 1.7 pp, low confidence
62	Intern-Decision-2B	86.7%	estimated ± 1.7 pp, low confidence
63	JPT-4B	86.5%	estimated ± 1.4 pp, low confidence
64	decider-2b	86.4%	estimated ± 1.7 pp, low confidence
65	Intern-Decision-4B	86.4%	estimated ± 1.4 pp, low confidence
66	Jev-Style 0.8B v3	86.4%	estimated ± 4.6 pp, low confidence
67	NeoHorse-Jev-4B	86.3%	estimated ± 1.4 pp, low confidence
68	Von 1.3	86.2%	estimated ± 4.6 pp, low confidence
69	JevK5 Lite	86.2%	estimated ± 4.6 pp, low confidence
70	Intelif Qwen3-4B	86.0%	estimated ± 1.4 pp, low confidence
71	GLiNER2.5-Decide	85.9%	estimated ± 4.6 pp, low confidence
72	jeff	85.7%	estimated ± 4.6 pp, low confidence
73	Lavoir	85.7%	estimated ± 4.6 pp, low confidence
74	Kev 0.8B r15	85.6%	estimated ± 4.6 pp, low confidence
75	razorback16 openjev diffusiongemma (NVFP4, vLLM)	85.6%	estimated ± 1.4 pp, low confidence
76	Kas 4B	85.6%	estimated ± 1.4 pp, low confidence
77	Winnow-E4B	85.5%	estimated ± 1.4 pp, low confidence
78	Bekko System One v0 17M	85.3%	estimated ± 4.6 pp, low confidence
79	Bespoke Nimble 9B v2	85.3%	estimated ± 1.4 pp, low confidence
80	JevK5 v0.3	85.3%	estimated ± 1.4 pp, low confidence
81	Sieve 2B	85.2%	estimated ± 1.7 pp, low confidence
82	djev	85.2%	estimated ± 1.4 pp, low confidence
83	vLLM-SR Decision 2.0 Eos 0.8B	84.8%	estimated ± 1.7 pp, low confidence
84	Bekko System One v0 400M	84.8%	estimated ± 1.7 pp, low confidence
85	JPT-0.8B	84.7%	estimated ± 1.7 pp, low confidence
86	Ryotide Qwen v0.3.1	84.7%	estimated ± 1.4 pp, low confidence
87	Kev 4B r10	84.0%	estimated ± 1.4 pp, low confidence
88	RSI-Jev v5.0 VL 3B	83.7%	estimated ± 1.4 pp, low confidence
89	Gevva E4B	83.2%	estimated ± 1.4 pp, low confidence
90	decider-35b-a3b	83.1%	estimated ± 1.4 pp, low confidence
91	Jobe	83.1%	estimated ± 1.4 pp, low confidence
92	Decider 4B	82.8%	estimated ± 1.4 pp, low confidence
93	openvons	82.5%	estimated ± 1.4 pp, low confidence
94	vLLM-SR Decision 2.0 Sol 2B	82.3%	estimated ± 1.4 pp, low confidence
95	Solomon v1.1	81.8%	estimated ± 1.4 pp, low confidence
96	Gevva E2B	81.5%	estimated ± 1.4 pp, low confidence
97	EXAONE-4.0-1.2B-JEV v0.3	81.1%	estimated ± 1.4 pp, low confidence
98	Tev1-4B-experimental	81.0%	estimated ± 1.4 pp, low confidence
99	SemIf	78.3%	estimated ± 1.4 pp, low confidence
100	this-that 1.2	77.6%	estimated ± 1.4 pp, low confidence
101	metask-jev-4b	77.6%	estimated ± 1.4 pp, low confidence

WildGuardTest (Decision Index) leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General