General

Decision Index Arts 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 Arts is 62.7% 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	62.7%	measured
2	elephant-1	56.7%	measured
3	d1-ultra	56.4%	estimated ± 2.0 pp, medium confidence
4	Perplexity Decider v1.2 27B	56.4%	measured
5	Drex v2.0	55.2%	measured
6	Blink v0.3 26B-A4B (NVFP4)	52.3%	measured
7	Torchcast Decision 27B	52.2%	measured
8	Surogate Rune 26B-A4B v3	52.0%	measured
9	Celeris-1 Decision	51.1%	estimated ± 2.0 pp, medium confidence
10	GLiDE	51.1%	estimated ± 2.0 pp, medium confidence
11	Drex v1.5	50.1%	measured
12	Microsoft-Decision-1	50.0%	estimated ± 2.0 pp, medium confidence
13	GEV-26B-Decide	48.6%	measured
14	Mercury Decide	48.5%	estimated ± 2.0 pp, medium confidence
15	Xor 1.2	48.1%	measured
16	Jev 1.14	47.7%	measured
17	d1	47.4%	estimated ± 2.0 pp, medium confidence
18	Bespoke Nimble 9B v3	47.2%	measured
19	Eikos-27B	47.1%	measured
20	JEV-27B	46.9%	measured
21	vLLM-SR Decision 2.0 Vega 27B	45.6%	measured
22	simple-jev · Qwen3.8-27B	45.3%	measured
23	JADE	45.2%	measured
24	Kev 27B	45.1%	measured
25	Quyet-1.0-Large	45.1%	measured
26	Perplexity Decider Flash v1.2 4B	45.0%	measured
27	Clef	44.7%	measured
28	deck31b (Gemma 4 31B)	44.3%	measured
29	Decider chat · Gemma-4-31B	44.3%	measured
30	decider-35b-a3b	42.9%	measured
31	Decider chat · Qwen3.6-27B	42.9%	measured
32	JPT-35B-A3B	41.4%	measured
33	JoshuaSP diffusiongemma (open-jev)	40.9%	measured
34	decisio 0.4.0 (Qwen3.6-35B-A3B)	40.2%	measured
35	lev	39.4%	measured
36	Clef-Flash	39.3%	measured
37	decisio 0.4.0 (Gemma 4 12B)	38.7%	measured
38	Lev1 3.7B	38.7%	measured
39	Clef-Omni	37.8%	measured
40	ezjev 4B s2	37.8%	measured
41	Jevfire	37.6%	measured
42	vLLM-SR Decision 2.0 Lux 9B	37.6%	measured
43	djev	37.4%	measured
44	Hopper 12B (trained)	37.3%	measured
45	razorback16 openjev diffusiongemma (NVFP4, vLLM)	35.9%	measured
46	Ryotide Qwen v0.3.1	35.7%	measured
47	JevK5 9B v0.3.3	35.1%	measured
48	Jobe	35.1%	measured
49	Winnow-12B Q8	35.1%	measured
50	JEV-9B	34.6%	measured
51	Decider 4B	34.5%	measured
52	Bespoke Nimble 9B v2	34.2%	measured
53	Hopper (G) 1.2	34.0%	measured
54	JevK5 v0.3	33.5%	measured
55	JPT-9B	33.2%	measured
56	mmastrac diffusiongemma (vLLM PR 57250)	33.2%	measured
57	SemIf	33.0%	measured
58	Intern-Decision-4B	32.9%	measured
59	openvons	32.7%	measured
60	mini-jev	31.9%	measured
61	vLLM-SR Decision 2.0 Nox 4B	31.9%	measured
62	Jev-Omni	31.9%	measured
63	Kas 4B	31.6%	measured
64	Solomon v1.1	30.7%	measured
65	JPT-4B	30.6%	measured
66	Sieve 9B	29.7%	measured
67	Gevva E4B	28.3%	measured
68	Winnow-E4B	28.2%	measured
69	Tev1-4B-experimental	27.3%	measured
70	Intelif Qwen3-4B	27.2%	measured
71	Bekko System One v0 400M	24.6%	measured
72	decider-2b	24.4%	measured
73	Sieve 2B	23.6%	measured
74	Kev 4B r10	23.0%	measured
75	metask-jev-4b	22.2%	measured
76	RSI-Jev v5.0 VL 3B	22.1%	measured
77	Intern-Decision-2B	21.4%	measured
78	Kev 9B v2	21.3%	measured
79	open-jev (pngwn)	20.3%	measured
80	this-that 1.2	19.1%	measured
81	Jevstral 8B	17.6%	measured
82	Kodiak v0.2 1B (accuracy)	17.3%	measured
83	Bekko System One v0 68M	16.9%	measured
84	vLLM-SR Decision 2.0 Sol 2B	16.3%	measured
85	EXAONE-4.0-1.2B-JEV v0.3	16.2%	measured
86	NeoHorse-Jev-4B	15.5%	measured
87	Qwen-2.5-1B-RLCD	13.2%	measured
88	Gevva E2B	13.0%	measured
89	LFM2.5-2.6B-RLCD	11.4%	measured
90	vLLM-SR Decision 2.0 Eos 0.8B	9.6%	measured
91	Sifr 0.8B v3.1	8.7%	measured
92	Bosun v3.1 1.7B	8.5%	measured
93	Intern-Decision-0.8B	8.2%	measured
94	vLLM-SR Decision 2.0 Kai 0.6B	8.2%	measured
95	Von 1.3	8.0%	measured
96	Jev-Style 0.8B v3	7.9%	measured
97	JPT-0.8B	7.8%	measured
98	system-one-gemma	7.3%	measured
99	Laya	7.0%	measured
100	Bosun v3.1 0.6B	6.6%	measured
101	JevK5 Lite	5.8%	measured
102	Tev1-0.8B-experimental	5.7%	measured
103	GLiNER2.5-Decide	5.1%	measured
104	Lavoir	4.8%	measured
105	Kev 0.8B r15	4.7%	measured
106	LFM2.5-350M-RLCD	4.3%	measured
107	Bekko System One v0 17M	3.9%	measured
108	DM-JEPA 1.1	3.6%	measured
109	Julia 1	2.4%	measured
110	CLM-v0.1-8B	1.5%	measured
111	Zev	1.1%	measured
112	jeff	1.0%	measured
113	Verdict	0.0%	measured

Decision Index Arts leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General