General

Decision Index Knowledge 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 Decision Index Knowledge is 52.1% by mammoth-1. 6 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	mammoth-1	52.1%	measured
2	Quyet-1.0-Large	47.0%	measured
3	Jev 1.14	46.6%	measured
4	Perplexity Decider v1.2 27B	46.0%	measured
5	elephant-1	45.8%	measured
6	deck31b (Gemma 4 31B)	44.8%	measured
7	Decider chat · Gemma-4-31B	44.8%	measured
8	Surogate Rune 26B-A4B v3	42.0%	measured
9	d1-ultra	41.4%	estimated ± 3.6 pp, medium confidence
10	Blink v0.3 26B-A4B (NVFP4)	40.7%	measured
11	JEV-27B	39.0%	measured
12	Torchcast Decision 27B	38.9%	measured
13	Celeris-1 Decision	38.8%	estimated ± 3.6 pp, medium confidence
14	GEV-26B-Decide	38.7%	measured
15	Kev 27B	38.3%	measured
16	Decider chat · Qwen3.6-27B	38.0%	measured
17	vLLM-SR Decision 2.0 Vega 27B	37.8%	measured
18	Eikos-27B	37.6%	measured
19	JADE	36.9%	measured
20	Clef	36.3%	measured
21	d1	36.0%	estimated ± 3.6 pp, medium confidence
22	Drex v2.0	36.0%	measured
23	JPT-35B-A3B	35.8%	measured
24	Xor 1.2	35.8%	measured
25	Microsoft-Decision-1	35.1%	estimated ± 3.6 pp, medium confidence
26	GLiDE	34.5%	estimated ± 3.6 pp, medium confidence
27	Clef-Omni	34.1%	measured
28	simple-jev · Qwen3.8-27B	33.9%	measured
29	Mercury Decide	33.9%	estimated ± 3.6 pp, medium confidence
30	Bespoke Nimble 9B v3	32.9%	measured
31	JoshuaSP diffusiongemma (open-jev)	32.7%	measured
32	Clef-Flash	32.5%	measured
33	JevK5 9B v0.3.3	32.2%	measured
34	decider-35b-a3b	32.0%	measured
35	Perplexity Decider Flash v1.2 4B	32.0%	measured
36	decisio 0.4.0 (Qwen3.6-35B-A3B)	31.2%	measured
37	Drex v1.5	31.2%	measured
38	Winnow-12B Q8	31.2%	measured
39	vLLM-SR Decision 2.0 Lux 9B	30.3%	measured
40	decisio 0.4.0 (Gemma 4 12B)	30.0%	measured
41	JEV-9B	29.8%	measured
42	JPT-9B	29.1%	measured
43	Kev 9B v2	28.6%	measured
44	ezjev 4B s2	28.5%	measured
45	vLLM-SR Decision 2.0 Nox 4B	28.4%	measured
46	Hopper 12B (trained)	27.8%	measured
47	JPT-4B	27.2%	measured
48	Bespoke Nimble 9B v2	26.6%	measured
49	Hopper (G) 1.2	26.5%	measured
50	Decider 4B	26.4%	measured
51	Kas 4B	26.2%	measured
52	JevK5 v0.3	25.8%	measured
53	Sieve 9B	25.2%	measured
54	Jevfire	24.5%	measured
55	open-jev (pngwn)	23.9%	measured
56	Jev-Omni	23.4%	measured
57	Kev 4B r10	23.3%	measured
58	NeoHorse-Jev-4B	23.3%	measured
59	mmastrac diffusiongemma (vLLM PR 57250)	23.2%	measured
60	Intern-Decision-4B	22.9%	measured
61	razorback16 openjev diffusiongemma (NVFP4, vLLM)	22.9%	measured
62	Tev1-4B-experimental	22.8%	measured
63	Jevstral 8B	22.5%	measured
64	djev	22.4%	measured
65	RSI-Jev v5.0 VL 3B	21.8%	measured
66	Solomon v1.1	21.4%	measured
67	metask-jev-4b	21.4%	measured
68	lev	20.8%	measured
69	Intelif Qwen3-4B	20.3%	measured
70	Lev1 3.7B	20.1%	measured
71	Winnow-E4B	20.1%	measured
72	openvons	19.9%	measured
73	Ryotide Qwen v0.3.1	18.2%	measured
74	mini-jev	18.1%	measured
75	Gevva E4B	16.7%	measured
76	Jobe	16.2%	measured
77	decider-2b	15.4%	measured
78	SemIf	15.3%	measured
79	EXAONE-4.0-1.2B-JEV v0.3	15.2%	measured
80	this-that 1.2	14.3%	measured
81	Gevva E2B	13.9%	measured
82	vLLM-SR Decision 2.0 Sol 2B	13.0%	measured
83	Sieve 2B	12.4%	measured
84	Intern-Decision-2B	9.5%	measured
85	Kodiak v0.2 1B (accuracy)	9.4%	measured
86	Sifr 0.8B v3.1	8.0%	measured
87	Bekko System One v0 400M	7.9%	measured
88	vLLM-SR Decision 2.0 Eos 0.8B	7.6%	measured
89	Kev 0.8B r15	7.4%	measured
90	JPT-0.8B	6.6%	measured
91	LFM2.5-2.6B-RLCD	6.6%	measured
92	Intern-Decision-0.8B	6.5%	measured
93	Von 1.3	6.2%	measured
94	vLLM-SR Decision 2.0 Kai 0.6B	6.0%	measured
95	JevK5 Lite	4.7%	measured
96	Tev1-0.8B-experimental	4.4%	measured
97	Julia 1	4.1%	measured
98	Bosun v3.1 1.7B	3.0%	measured
99	Jev-Style 0.8B v3	3.0%	measured
100	Lavoir	2.6%	measured
101	system-one-gemma	2.3%	measured
102	Verdict	2.1%	measured
103	GLiNER2.5-Decide	2.0%	measured
104	Qwen-2.5-1B-RLCD	1.6%	measured
105	Laya	1.4%	measured
106	Bekko System One v0 68M	0.9%	measured
107	jeff	0.9%	measured
108	Bosun v3.1 0.6B	0.4%	measured
109	DM-JEPA 1.1	0.4%	measured
110	CLM-v0.1-8B	0.1%	measured
111	Bekko System One v0 17M	0.0%	measured
112	LFM2.5-350M-RLCD	0.0%	measured
113	Zev	0.0%	measured

Decision Index Knowledge leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General