General

Decision Index Retrieval 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 Retrieval is 69.4% by Perplexity Decider v1.2 27B. 6 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Perplexity Decider v1.2 27B	69.4%	measured
2	d1-ultra	68.4%	estimated ± 7.6 pp, low confidence
3	mammoth-1	66.2%	measured
4	elephant-1	65.1%	measured
5	deck31b (Gemma 4 31B)	64.8%	measured
6	Jev 1.14	64.8%	measured
7	Celeris-1 Decision	64.0%	estimated ± 7.6 pp, low confidence
8	GLiDE	64.0%	estimated ± 7.6 pp, low confidence
9	simple-jev · Qwen3.8-27B	63.5%	measured
10	Decider chat · Gemma-4-31B	63.4%	measured
11	GEV-26B-Decide	63.3%	measured
12	Quyet-1.0-Large	63.2%	measured
13	Torchcast Decision 27B	63.2%	measured
14	Blink v0.3 26B-A4B (NVFP4)	63.1%	measured
15	Microsoft-Decision-1	63.0%	estimated ± 7.6 pp, low confidence
16	Surogate Rune 26B-A4B v3	62.3%	measured
17	Perplexity Decider Flash v1.2 4B	62.2%	measured
18	Mercury Decide	61.6%	estimated ± 7.6 pp, low confidence
19	vLLM-SR Decision 2.0 Vega 27B	61.1%	measured
20	d1	60.6%	estimated ± 7.6 pp, low confidence
21	Clef	59.8%	measured
22	Eikos-27B	59.3%	measured
23	Drex v1.5	59.2%	measured
24	JEV-27B	58.5%	measured
25	Drex v2.0	58.5%	measured
26	Xor 1.2	57.9%	measured
27	Clef-Omni	56.5%	measured
28	decisio 0.4.0 (Gemma 4 12B)	56.5%	measured
29	JADE	56.0%	measured
30	Kev 27B	55.6%	measured
31	JPT-35B-A3B	55.1%	measured
32	Winnow-12B Q8	54.6%	measured
33	vLLM-SR Decision 2.0 Lux 9B	54.3%	measured
34	JoshuaSP diffusiongemma (open-jev)	54.2%	measured
35	decisio 0.4.0 (Qwen3.6-35B-A3B)	54.0%	measured
36	Decider chat · Qwen3.6-27B	53.8%	measured
37	Jevfire	53.7%	measured
38	Bespoke Nimble 9B v3	52.9%	measured
39	Hopper 12B (trained)	52.9%	measured
40	vLLM-SR Decision 2.0 Nox 4B	52.3%	measured
41	decider-35b-a3b	52.1%	measured
42	ezjev 4B s2	52.0%	measured
43	JevK5 9B v0.3.3	52.0%	measured
44	JEV-9B	48.4%	measured
45	RSI-Jev v5.0 VL 3B	48.3%	measured
46	Kev 9B v2	47.4%	measured
47	Hopper (G) 1.2	46.6%	measured
48	Ryotide Qwen v0.3.1	45.8%	measured
49	Sieve 9B	45.8%	measured
50	JevK5 v0.3	45.6%	measured
51	djev	44.8%	measured
52	NeoHorse-Jev-4B	44.7%	measured
53	Lev1 3.7B	44.0%	measured
54	JPT-9B	42.3%	measured
55	lev	42.3%	measured
56	Decider 4B	42.2%	measured
57	Kev 4B r10	41.7%	measured
58	Clef-Flash	40.8%	measured
59	Kas 4B	40.8%	measured
60	Winnow-E4B	40.8%	measured
61	JPT-4B	40.4%	measured
62	vLLM-SR Decision 2.0 Sol 2B	40.4%	measured
63	Intern-Decision-4B	39.5%	measured
64	Jev-Omni	39.3%	measured
65	Gevva E2B	38.5%	measured
66	Bosun v3.1 1.7B	37.9%	measured
67	this-that 1.2	37.9%	measured
68	Bekko System One v0 400M	37.7%	measured
69	razorback16 openjev diffusiongemma (NVFP4, vLLM)	37.6%	measured
70	Jobe	37.5%	measured
71	decider-2b	36.7%	measured
72	Intelif Qwen3-4B	36.6%	measured
73	Jevstral 8B	36.6%	measured
74	Bespoke Nimble 9B v2	36.0%	measured
75	vLLM-SR Decision 2.0 Eos 0.8B	33.1%	measured
76	Gevva E4B	32.9%	measured
77	EXAONE-4.0-1.2B-JEV v0.3	31.2%	measured
78	openvons	30.0%	measured
79	Kodiak v0.2 1B (accuracy)	29.9%	measured
80	Bekko System One v0 68M	28.5%	measured
81	Bosun v3.1 0.6B	28.1%	measured
82	Sieve 2B	28.0%	measured
83	Solomon v1.1	26.9%	measured
84	vLLM-SR Decision 2.0 Kai 0.6B	26.8%	measured
85	Sifr 0.8B v3.1	26.0%	measured
86	open-jev (pngwn)	24.0%	measured
87	mmastrac diffusiongemma (vLLM PR 57250)	23.0%	measured
88	Jev-Style 0.8B v3	21.7%	measured
89	Von 1.3	20.9%	measured
90	JevK5 Lite	20.3%	measured
91	Tev1-4B-experimental	18.5%	measured
92	GLiNER2.5-Decide	18.3%	measured
93	SemIf	17.8%	measured
94	Intern-Decision-2B	17.7%	measured
95	jeff	17.2%	measured
96	Lavoir	17.1%	measured
97	Kev 0.8B r15	16.7%	measured
98	JPT-0.8B	14.6%	measured
99	Bekko System One v0 17M	14.5%	measured
100	metask-jev-4b	14.1%	measured
101	mini-jev	10.0%	measured
102	Laya	9.2%	measured
103	system-one-gemma	8.8%	measured
104	Intern-Decision-0.8B	8.5%	measured
105	LFM2.5-2.6B-RLCD	4.9%	measured
106	CLM-v0.1-8B	4.6%	measured
107	DM-JEPA 1.1	4.3%	measured
108	Julia 1	3.5%	measured
109	Tev1-0.8B-experimental	3.4%	measured
110	Qwen-2.5-1B-RLCD	1.0%	measured
111	LFM2.5-350M-RLCD	0.0%	measured
112	Verdict	0.0%	measured
113	Zev	0.0%	measured

Decision Index Retrieval leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General