S1 Decision Index0.4.1
247 open reproductions and similar decision models, re-tested by the Decision Index and ranked by its composite score, with Jev 1.14 as the reference. Public benchmarks are shown, never scored.
Decision Index 0.4.1 · data updated Oct 29, 2026, 2:04 PM · synced Oct 30, 2026, 3:00 AM (daily) · Source: Decision Index on Hugging Face
Max paramsAll
<100M1B5B10B20B35B+
Reference
60.2
Jev 1.14
#1 · tied
71.9
KnowLine 31B Gen1
71.5
Fastino GLiDE 1.2
#3 · tied
71.1
Sieve 31B v3
70.5
Turion Jev 31B
#5 · tied
70.2
vLLM-SR d3-ultra (Decision 3.0, 27B)
69.8
Jeffan-31b-1710
#7 · tied
68.6
H2O Lightning 31B v1.1
68.5
WeMM-JEV-27B
68.4
baselabs mammoth-1 (27B)
68.2
Ainvest-Decision-27B v1.0
68.1
oenogarum (Gemma 4 31B)
68.0
AutoTrust-JEV-27B v1.1
68.0
Perplexity Decider v1.2 (27B)
67.9
StartLux-Decision v1.1 27B
#15 · tied
67.6
AJev lm6q (Gemma 4 31B)
67.6
FlexDM · Gemma-4-31B
67.6
MISA AI 2.0 Decision Plus · oak 1.1 (31B)
67.4
Lychee-27B
67.3
vLLM-SR d3-turbo v1.1 (27B)
67.2
vLLM-SR d3 (27B)
67.0
allec-oavg (Gemma 4 31B)
66.9
VisionJEV 1.0.0 (27B)
#23 · tied
66.7
Aleria Judge 1.1.1 (Gemma 4 31B)
66.6
Bespoke Nimble v4 Max (27B)
66.6
Nace.AI Drex Max v1 (27B)
66.5
AnyJev Tacitus 31B
66.0
Decider 31B (NVFP4)
#28 · tied
65.8
KnowLine 35B-A3B Gen1.5
65.7
OneShot 27B
65.3
vera-large
#31 · tied
64.9
Bobcat 1.3 (31B)
64.8
Amos 1 Large (27B)
64.8
Matilda Decision v1
64.6
KnowLine 26B-A4B Gen2.5
64.6
jev-brain · Gemma-4-31B
64.4
frontier-infra Jebadiah 27B v2.1 (t1e)
#37 · tied
63.7
AnyJev L0 · Gemma-4-31B
63.1
deck31b (Gemma 4 31B)
63.0
Darwin-27B-ZTC
#40
62.7
decisio 0.6.0 (Gemma 4 31B)
#41 · tied
61.7
Cerno 27B
61.6
Decider chat · Gemma-4-31B
61.5
OneJev1.6-27B
61.5
Torchcast Decision 27B
61.1
Qwen Decision 27B noncausal LoRA
#46 · tied
60.8
Quyet-1.0-Large
60.3
baselabs elephant-1 (9B)
60.3
garum (Gemma 4 31B)
60.1
Standard One 35B SH (v1)
60.0
Lumma-fev-27B
#51 · tied
59.9
Sieve 27B v2
59.8
mogan-decision-31B-it
59.7
linne-fern 1.0 (27B)
59.6
Kev 27B
59.4
Xor 26B-A4B v1.2
59.2
Jev 1.13
59.1
span-02 (Gemma 4 26B-A4B)
#58 · tied
59.0
garas (Gemma 4 31B)
58.8
Eikos-27B
58.8
Shisa DE-2.1 (repeat)
58.8
reflex Qwen3.8-27B-FP8 (pmscr)
58.6
RSI-Jev v6.1 VL 27B (effort auto)
58.4
KnowLine 9B Gen1
#64 · tied
58.1
StartLux-Decision v1.1 9B
58.0
vLLM-SR Decision 2.0 Vega 27B
57.6
JEV-35B
57.4
JEV-27B
#68 · tied
57.0
simple-jev · Qwen3.8-27B (featherless)
56.6
Tern-35B-A3B
56.4
Surogate Rune 26B-A4B v3
56.2
SynACK Decide 26B-A4B v7
#72 · tied
56.0
KnowLine 4B Gen5.5
55.9
Blink v0.3 26B-A4B (NVFP4)
55.6
JADE
55.4
Wald-26B-A4B
55.3
AJev lora1 (Gemma 4 26B-A4B)
#77 · tied
54.9
Cloudflare clef
54.8
StartLux-Decision v1.1 4B
54.7
Decider chat · Qwen3.6-27B
54.3
Perplexity Decider Flash v1.2 (4B)
54.3
vLLM-SR d3-flash (9B)
54.1
JPT-35B-A3B
#83 · tied
53.6
Xor 1.2
53.5
GEV-26B-Decide
53.1
OneJev1.6-9B
#86 · tied
52.8
Nace.AI Drex v2.0 (35B-A3B)
52.6
muria (Gemma 4 12B)
52.2
Jeffan-4B-2010
52.0
Verdin-4B v1.1
#90 · tied
51.1
Bespoke Nimble 9B v3
51.0
Senova-4B v2
50.9
frontier-infra Jebadiah 9B v2.1 (t1e)
50.8
decisio 0.4.0 (Qwen3.6-35B-A3B)
#94 · tied
50.0
vLLM-SR d3-mini (4B)
49.5
JoshuaSP diffusiongemma (open-jev)
#96 · tied
48.9
AJev lora7 (Gemma 4 12B)
48.9
OneJev1.6-4B
48.8
decisio 0.4.0 (Gemma 4 12B)
#99 · tied
47.7
Mica v0.5 4B
47.5
PurpleMIST-Flash-1.0
47.3
Cerno 9B v1.1
47.0
Sieve 9B v2
#103 · tied
46.6
Decider 35B-A3B
46.6
Winnow-12B
46.3
Seb-9B v2
46.0
vLLM-SR Decision 2.0 Lux 9B
#107 · tied
45.8
Hopper 12B (trained)
45.8
Nace.AI Drex v1.5
45.6
Trio-Spark DI (9B)
45.5
CoCo-Decision-4B 2.4
45.0
Cloudflare clef-omni (30B-A3B)
#112 · tied
44.9
JevK5 9B v0.3.3
44.4
Cloudflare clef-flash
44.4
Jevfire
44.3
RSI-Jev v6.1 VL 4B
44.2
Jegue 0.1 (9B)
44.1
JPT-9B
44.1
jiwo 4B v2
#119 · tied
43.9
ezjev 4B s2
43.8
Jev-Omni
43.2
Solomon v1.1
#122 · tied
42.9
Standard One 8B SH (v4)
42.7
Amber Hybrid 4B
42.5
Saif 4B
42.2
JEV-9B
42.1
Bespoke Nimble 9B v2
#127 · tied
41.7
Sieve 9B
41.2
Cerno 4B v1.1
41.0
Sieve 4B
#130 · tied
40.9
Kev 9B v2
40.9
Nari Decision 4B v3.2
40.8
Aplomb 1
40.8
vLLM-SR Decision 2.0 Nox 4B
40.2
Hopper (G) 1.2
40.1
StartLux-Decision v1.1 2B
#136 · tied
39.8
FlyMy.AI Decision 4B v1.6
39.8
Standard One 3B SH (v4)
39.6
JPT-4B
39.5
Amber 4B
#140 · tied
38.7
Intern-Decision-4B
38.7
RSI-Jev v5.0 VL 3B
38.2
djev
#143 · tied
37.9
Valen-4B
37.6
razorback16 openjev diffusiongemma (NVFP4, vLLM)
37.3
NeoHorse-Jev-4B
#146 · tied
36.6
JevK5 v0.3
36.3
Decider 4B
35.9
baselabs joey-1 (2B)
#149 · tied
35.5
lev
35.4
Lev1 3.7B
34.9
Kev 4B r10
#152 · tied
34.5
Jev-Style 2B v4
34.5
Sieve 2B v3
34.4
Cam1
34.4
LatentMesh Blitz 2B v1
34.4
Saracura PT-BR 4B
34.3
mmastrac diffusiongemma (vLLM PR 57250)
34.3
vLLM-SR d3-nano (2B)
34.0
Kas 4B
33.7
Tev1-4B-experimental
#161 · tied
32.7
Winnow-E4B
32.6
Aurelis-2B
32.2
Jevstral 8B
32.2
Metask-Jev-4B
32.0
Ryotide Qwen v0.3.1
#166 · tied
31.3
Intelif Qwen3-4B
31.2
GLiClass v4 edge (0.4B)
31.1
LiquidAI d1-3B
31.1
open-jev (pngwn)
30.6
vera-core-2
#171 · tied
30.4
Jobe
29.6
OneJev1.6-0.8B
#173 · tied
29.4
crystal-0.8b
29.3
Crivo-1B
29.3
vLLM-SR vela3-fast (Vela 3.0, 0.4B)
29.2
Gevva E4B
28.9
StartLux-Decision v1.1 0.8B
28.8
openvons
#179 · tied
26.7
SemIf
26.5
Saina Helm 2 Omni (0.8B)
26.2
vLLM-SR d3-lite (0.8B)
26.0
Amber 0.4B
#183 · tied
25.8
jiwo 0.8B v2
25.3
Saina Helm 2.0 v2
#185 · tied
24.9
this-that 1.2
24.3
Gevva E2B
#187 · tied
23.7
Crivo-400M
23.7
Decider 2B
23.6
jdi-ptr14-0.8b
23.6
vLLM-SR Decision 2.0 Sol 2B
Jev 1.14 60.2
Composite score as published by the source. Hover a bar for its parts and the source’s own rank.
Params vs scorePareto frontier
Latency vs scoreSystem1 is about speed
How it is scored
Scores, ranks and ties come straight from the Decision Index: private tests of the same skills as public benchmarks (S), private tasks from new domains (N), and public benchmarks (P) shown for comparison only.
SPrivate · same skills
Tests the same abilities as 37 public benchmarks, with questions that were never published.
NPrivate · new domains
Real decision tasks from 8 industries, written by verified practitioners.
PPublic benchmarks
Shown in grey to expose the public-minus-private gap; a large gap suggests overfitting.
Ties: models within the 95% confidence interval of their group leader (≈0.8 points) share a rank.
Skill-area rankings
The source scores five skill areas. Each has its own leader — click one to re-rank the chart above.
Head to head
Pick any two models and compare both private parts and all 8 industry sub-scores.
VS
67.2Private · same skills59.3
62.0Private · new domains48.3
59.4📚 Knowledge & reasoning57.1
72.0💬 Language61.2
73.0🔎 Retrieval & intent63.5
84.7🛠 Tool use73.8
60.5🎨 Arts & taste39.2
Full results
Click a header to sort. Darker industry cells mean higher scores.
| # | Model | Method | Params | Score ↓ | Same skills | New domains | Public | Calib. error ↓ | Latency ms | Source rank | 📚 Knowledge & reasoning | 💬 Language | 🔎 Retrieval & intent | 🛠 Tool use | 🎨 Arts & taste |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1= | KnowLine 31B Gen1 | LoRA | 31.27B | 71.9 | 67.2 | 62.0 | 70.1 | 3.2 | 204 | #1= | 59.4 | 72.0 | 73.0 | 84.7 | 60.5 |
| 1= | Fastino GLiDE 1.2 | Full fine-tune | 27.8B | 71.5 | 68.6 | 59.7 | 72.0 | 2.0 | 93 | #2= | 60.6 | 75.5 | 72.2 | 88.0 | 62.6 |
| 3= | Sieve 31B v3 | Full fine-tune | 31.27B | 71.1 | 68.5 | 58.9 | 71.7 | 5.4 | 232 | #3= | 61.2 | 71.6 | 76.1 | 86.9 | 62.3 |
| 3= | Turion Jev 31B | LoRA | 31.27B | 70.5 | 65.3 | 61.5 | 66.2 | 1.9 | 144 | #4= | 55.4 | 69.9 | 71.2 | 78.7 | 51.3 |
| 5= | vLLM-SR d3-ultra (Decision 3.0, 27B) | Full fine-tune | 26.09B | 70.2 | 66.8 | 59.2 | 71.6 | 6.7 | 201 | #5= | 62.3 | 72.4 | 71.8 | 87.8 | 63.5 |
| 5= | Jeffan-31b-1710 | Full fine-tune | 31.27B | 69.8 | 66.2 | 59.1 | 73.1 | 4.7 | 121 | #6= | 63.2 | 76.0 | 72.2 | 89.4 | 63.6 |
| 7= | H2O Lightning 31B v1.1 | Full fine-tune | 31.27B | 68.6 | 65.6 | 57.5 | 67.8 | 7.4 | 100 | #7= | 56.5 | 69.6 | 66.9 | 86.3 | 59.7 |
| 7= | WeMM-JEV-27B | Full fine-tune | 26.09B | 68.5 | 65.4 | 57.5 | 69.4 | 5.1 | 98 | #8= | 59.0 | 74.1 | 70.1 | 86.0 | 52.4 |
| 7= | baselabs mammoth-1 (27B) | Full fine-tune | 27.36B | 68.4 | 66.3 | 56.3 | 75.3 | 4.1 | 220 | #9= | 68.0 | 76.1 | 77.4 | 88.6 | 64.1 |
| 7= | Ainvest-Decision-27B v1.0 | Full fine-tune | 26.09B | 68.2 | 64.6 | 57.9 | 71.9 | 4.2 | 99 | #10= | 57.9 | 74.7 | 75.5 | 87.1 | 65.7 |
| 7= | oenogarum (Gemma 4 31B) | Full fine-tune | 31.27B | 68.1 | 62.8 | 59.8 | 62.8 | 4.0 | 259 | #11= | 52.3 | 66.0 | 64.6 | 77.3 | 51.5 |
| 7= | AutoTrust-JEV-27B v1.1 | Full fine-tune | 27.36B | 68.0 | 64.1 | 58.0 | 67.3 | 6.7 | 80 | #13= | 55.2 | 73.4 | 67.0 | 85.2 | 51.0 |
| 7= | Perplexity Decider v1.2 (27B) | Full fine-tune | 26.09B | 68.0 | 64.3 | 57.8 | 67.7 | 5.5 | 144 | #12= | 55.3 | 71.9 | 68.1 | 85.4 | 56.2 |
| 7= | StartLux-Decision v1.1 27B | Full fine-tune | 27.78B | 67.9 | 64.8 | 57.0 | 72.9 | 2.6 | 80 | #14= | 60.7 | 75.6 | 74.8 | 87.9 | 66.2 |
| 15= | AJev lm6q (Gemma 4 31B) | LoRA | 31.27B | 67.6 | 63.4 | 58.1 | 65.5 | 9.6 | 144 | #16= | 53.8 | 68.6 | 67.3 | 81.5 | 54.4 |
| 15= | FlexDM · Gemma-4-31B | Inference technique | 31.27B | 67.6 | 63.0 | 58.5 | 63.6 | 4.4 | 109 | #17= | 51.1 | 66.4 | 69.4 | 79.7 | 48.0 |
| 15= | MISA AI 2.0 Decision Plus · oak 1.1 (31B) | Full fine-tune | 31.27B | 67.6 | 62.6 | 59.2 | 68.2 | 7.2 | 113 | #15= | 54.4 | 70.9 | 71.8 | 85.7 | 57.5 |
| 15= | Lychee-27B | Full fine-tune | 26.09B | 67.4 | 62.5 | 58.7 | 62.8 | 10.8 | 97 | #18= | 51.8 | 66.2 | 63.9 | 80.8 | 47.0 |
| 15= | vLLM-SR d3-turbo v1.1 (27B) | Full fine-tune | 26.09B | 67.3 | 64.3 | 56.4 | 71.6 | 5.2 | 191 | #19= | 58.9 | 74.5 | 75.8 | 86.9 | 60.9 |
| 15= | vLLM-SR d3 (27B) | Full fine-tune | 26.09B | 67.2 | 62.8 | 57.9 | 65.5 | 4.3 | 97 | #20= | 55.3 | 68.0 | 66.9 | 82.6 | 51.0 |
| 15= | allec-oavg (Gemma 4 31B) | Full fine-tune | 31.27B | 67.0 | 62.4 | 58.0 | 62.0 | 4.7 | 257 | #21= | 50.7 | 65.2 | 64.8 | 77.6 | 49.0 |
| 15= | VisionJEV 1.0.0 (27B) | Full fine-tune | 26.09B | 66.9 | 62.8 | 57.4 | 66.5 | 7.4 | 90 | #22= | 55.8 | 71.4 | 68.0 | 82.7 | 48.7 |
| 23= | Aleria Judge 1.1.1 (Gemma 4 31B) | Inference technique | 31.27B | 66.7 | 62.4 | 57.4 | 63.5 | 8.9 | 117 | #23= | 51.0 | 67.9 | 66.2 | 79.0 | 50.3 |
| 23= | Bespoke Nimble v4 Max (27B) | Full fine-tune | 27.36B | 66.6 | 62.6 | 57.0 | 68.5 | 6.4 | 107 | #25= | 55.4 | 70.4 | 71.6 | 88.6 | 54.3 |
| 23= | Nace.AI Drex Max v1 (27B) | Full fine-tune | 26.09B | 66.6 | 63.1 | 56.5 | 67.5 | 5.7 | 104 | #24= | 57.1 | 72.2 | 68.8 | 81.9 | 53.4 |
| 23= | AnyJev Tacitus 31B | Full fine-tune | 31.27B | 66.5 | 62.5 | 56.9 | 65.5 | 2.3 | 95 | #26= | 52.8 | 70.3 | 67.6 | 81.3 | 53.2 |
| 23= | Decider 31B (NVFP4) | Full fine-tune | 32.68B | 66.0 | 61.8 | 56.9 | 64.3 | 2.4 | 72 | #27= | 50.3 | 66.7 | 65.5 | 84.2 | 55.2 |
| 28= | KnowLine 35B-A3B Gen1.5 | LoRA | 35.95B | 65.8 | 65.5 | 51.8 | 70.7 | 4.4 | 47 | #28= | 55.0 | 72.9 | 75.4 | 88.3 | 64.2 |
| 28= | OneShot 27B | Full fine-tune | 27.78B | 65.7 | 62.1 | 56.0 | 65.4 | 2.3 | 128 | #29= | 52.9 | 70.9 | 65.1 | 81.9 | 54.2 |
| 28= | vera-large | Full fine-tune | 25.63B | 65.3 | 61.6 | 55.6 | 67.8 | 5.1 | 160 | #30= | 51.2 | 75.7 | 69.5 | 83.0 | 58.5 |
| 31= | Bobcat 1.3 (31B) | LoRA | 31.27B | 64.9 | 59.9 | 56.8 | 59.3 | 10.2 | 64 | #31= | 48.0 | 61.0 | 65.5 | 76.1 | 40.9 |
| 31= | Amos 1 Large (27B) | Full fine-tune | 27.78B | 64.8 | 62.7 | 53.3 | 64.9 | 3.2 | 148 | #32= | 52.4 | 72.4 | 66.1 | 79.2 | 49.5 |
| 31= | Matilda Decision v1 | Full fine-tune | 26.09B | 64.8 | 60.3 | 56.1 | 65.4 | 8.6 | 92 | #33= | 51.1 | 71.3 | 67.6 | 84.6 | 47.4 |
| 31= | KnowLine 26B-A4B Gen2.5 | LoRA | 25.81B | 64.6 | 62.8 | 52.7 | 68.2 | 1.9 | 72 | #35= | 55.1 | 71.2 | 70.9 | 83.3 | 61.4 |
| 31= | jev-brain · Gemma-4-31B | Inference technique | 31.27B | 64.6 | 61.6 | 54.3 | 61.7 | 13.3 | 101 | #34= | 48.6 | 65.8 | 67.6 | 76.7 | 45.7 |
| 31= | frontier-infra Jebadiah 27B v2.1 (t1e) | LoRA | 27.78B | 64.4 | 59.4 | 56.6 | 60.5 | 5.6 | 303 | #36= | 48.2 | 65.5 | 59.9 | 79.3 | 45.8 |
| 37= | AnyJev L0 · Gemma-4-31B | Inference technique | 31.27B | 63.7 | 59.1 | 55.3 | 59.8 | 24.0 | 98 | #37= | 47.8 | 61.7 | 63.5 | 76.4 | 48.0 |
| 37= | deck31b (Gemma 4 31B) | Inference technique | 32.68B | 63.1 | 58.6 | 54.8 | 58.2 | 20.4 | 219 | #38= | 46.7 | 60.0 | 62.8 | 75.6 | 42.3 |
| 37= | Darwin-27B-ZTC | Full fine-tune | 25.62B | 63.0 | 58.7 | 54.6 | 62.6 | 4.5 | 104 | #39= | 44.7 | 68.5 | 60.9 | 86.7 | 53.3 |
| 40 | decisio 0.6.0 (Gemma 4 31B) | Inference technique | 31.27B | 62.7 | 58.5 | 54.2 | 58.1 | 14.2 | 54 | #40 | 47.2 | 59.8 | 62.7 | 75.0 | 41.4 |
| 41= | Cerno 27B | Full fine-tune | 26.9B | 61.7 | 59.1 | 51.4 | 65.2 | 7.2 | 54 | #41= | 49.0 | 71.1 | 65.9 | 86.7 | 50.9 |
| 41= | Decider chat · Gemma-4-31B | Inference technique | 32.68B | 61.6 | 56.9 | 54.0 | 57.8 | 4.7 | 109 | #42= | 45.5 | 59.5 | 63.1 | 75.6 | 42.2 |
| 41= | OneJev1.6-27B | Full fine-tune | 27.36B | 61.5 | 62.5 | 46.8 | 69.6 | 2.7 | 112 | #44= | 57.5 | 71.2 | 73.7 | 85.5 | 59.2 |
| 41= | Torchcast Decision 27B | LoRA | 27.8B | 61.5 | 58.1 | 52.2 | 65.1 | 2.4 | 100 | #43= | 46.7 | 72.0 | 67.5 | 84.6 | 54.4 |
| 41= | Qwen Decision 27B noncausal LoRA | LoRA | 27.78B | 61.1 | 57.7 | 52.0 | 57.7 | 5.1 | 127 | #45= | 42.8 | 62.3 | 57.8 | 80.2 | 43.1 |
| 46= | Quyet-1.0-Large | LoRA | 32.68B | 60.8 | 58.8 | 49.9 | 60.8 | 7.8 | 105 | #46= | 49.7 | 67.0 | 65.8 | 72.6 | 41.9 |
| 46= | baselabs elephant-1 (9B) | Full fine-tune | 9.41B | 60.3 | 62.4 | 44.4 | 72.2 | 5.8 | 114 | #48= | 60.6 | 73.8 | 76.2 | 88.0 | 61.6 |
| 46= | garum (Gemma 4 31B) | LoRA | 31.27B | 60.3 | 58.8 | 48.9 | 58.4 | 5.1 | 119 | #47= | 44.8 | 61.8 | 63.3 | 75.0 | 44.2 |
| — | Jev 1.14 | Hosted API | — | 60.2 | 59.3 | 48.3 | 60.7 | 8.0 | 196 | #49= | 57.1 | 61.2 | 63.5 | 73.8 | 39.2 |
| 46= | Standard One 35B SH (v1) | LoRA | 35.11B | 60.1 | 61.8 | 45.0 | 66.3 | 5.4 | 320 | #50= | 52.2 | 66.0 | 72.5 | 86.1 | 54.6 |
| 46= | Lumma-fev-27B | Full fine-tune | 26.09B | 60.0 | 56.6 | 51.0 | 57.2 | 6.8 | 100 | #51= | 42.5 | 60.8 | 57.7 | 78.8 | 44.8 |
| 51= | Sieve 27B v2 | LoRA | 27.78B | 59.9 | 60.0 | 46.6 | 68.0 | 5.3 | 106 | #52= | 56.8 | 66.0 | 73.4 | 85.8 | 58.9 |
| 51= | mogan-decision-31B-it | LoRA | 31.27B | 59.8 | 57.7 | 49.4 | 65.6 | 5.4 | 109 | #53= | 51.6 | 70.0 | 69.6 | 79.7 | 56.1 |
| 51= | linne-fern 1.0 (27B) | Full fine-tune | 27.78B | 59.7 | 56.5 | 50.4 | 58.9 | 7.9 | 108 | #54= | 44.9 | 62.8 | 62.4 | 79.0 | 41.5 |
| 51= | Kev 27B | Full fine-tune | 27.8B | 59.6 | 55.5 | 51.5 | 56.7 | 2.2 | 135 | #55= | 45.7 | 62.6 | 55.8 | 72.7 | 42.7 |
| 51= | Xor 26B-A4B v1.2 | LoRA | 25.81B | 59.4 | 58.7 | 47.3 | 59.5 | 6.8 | 110 | #56= | 45.5 | 63.2 | 64.8 | 72.2 | 51.9 |
| 51= | Jev 1.13 | Hosted API | — | 59.2 | 58.0 | 47.7 | 58.0 | 7.8 | 524 | #57= | 53.9 | 59.2 | 55.4 | 75.1 | 39.1 |
| 51= | span-02 (Gemma 4 26B-A4B) | Full fine-tune | 25.81B | 59.1 | 59.3 | 45.8 | 63.6 | 11.4 | 33 | #58= | 52.0 | 65.3 | 65.5 | 83.9 | 48.6 |
| 58= | garas (Gemma 4 31B) | LoRA | 31.27B | 59.0 | 56.6 | 48.9 | 56.6 | 4.9 | 117 | #59= | 43.3 | 58.7 | 61.9 | 73.8 | 43.6 |
| 58= | Eikos-27B | LoRA | 27.78B | 58.8 | 55.9 | 49.4 | 52.9 | 5.7 | 130 | #61= | 40.0 | 52.6 | 55.9 | 74.4 | 41.4 |
| 58= | Shisa DE-2.1 (repeat) | LoRA | 25.81B | 58.8 | 56.6 | 48.7 | 59.2 | 16.5 | 55 | #60= | 47.3 | 61.9 | 64.8 | 76.3 | 40.9 |
| 58= | reflex Qwen3.8-27B-FP8 (pmscr) | Inference technique | 27.78B | 58.8 | 57.4 | 47.6 | 57.3 | 4.5 | 149 | #62= | 37.3 | 62.8 | 64.2 | 77.2 | 45.2 |
| 58= | RSI-Jev v6.1 VL 27B (effort auto) | LoRA | 27.78B | 58.6 | 58.2 | 46.3 | 65.6 | 7.7 | 76 | #63= | 51.7 | 69.1 | 69.1 | 81.2 | 57.3 |
| 58= | KnowLine 9B Gen1 | LoRA | 9.65B | 58.4 | 60.8 | 42.7 | 65.7 | 2.8 | 128 | #64= | 47.6 | 69.1 | 71.9 | 82.7 | 59.8 |
| 64= | StartLux-Decision v1.1 9B | Full fine-tune | 9.65B | 58.1 | 61.3 | 41.4 | 70.0 | 4.6 | 39 | #65= | 53.4 | 73.1 | 74.3 | 88.2 | 63.1 |
| 64= | vLLM-SR Decision 2.0 Vega 27B | LoRA | 27.8B | 58.0 | 55.7 | 48.1 | 57.0 | 8.6 | 120 | #66= | 46.0 | 59.0 | 60.9 | 73.5 | 42.0 |
| 64= | JEV-35B | LoRA | 35.95B | 57.6 | 56.2 | 46.7 | 59.7 | 8.5 | 51 | #67= | 42.6 | 63.2 | 71.0 | 76.4 | 42.2 |
| 64= | JEV-27B | Full fine-tune | 27.8B | 57.4 | 54.2 | 48.6 | 53.6 | 7.6 | 102 | #68= | 41.4 | 56.4 | 53.7 | 73.5 | 41.6 |
| 68= | simple-jev · Qwen3.8-27B (featherless) | Inference technique | 27.78B | 57.0 | 55.5 | 46.4 | 56.1 | 11.3 | 373 | #69= | 36.9 | 61.2 | 63.2 | 76.2 | 40.8 |
| 68= | Tern-35B-A3B | Full fine-tune | 34.77B | 56.6 | 56.0 | 45.0 | 59.0 | 3.4 | 63 | #70= | 43.2 | 64.9 | 63.5 | 77.0 | 43.0 |
| 68= | Surogate Rune 26B-A4B v3 | Full fine-tune | 25.81B | 56.4 | 57.1 | 43.2 | 58.3 | 12.0 | 121 | #71= | 46.0 | 61.9 | 63.5 | 71.2 | 46.5 |
| 68= | SynACK Decide 26B-A4B v7 | LoRA | 25.81B | 56.2 | 56.3 | 43.8 | 59.7 | 6.9 | 129 | #72= | 43.7 | 63.5 | 62.4 | 79.1 | 50.1 |
| 72= | KnowLine 4B Gen5.5 | Full fine-tune | 4.66B | 56.0 | 61.2 | 37.3 | 66.8 | 4.7 | 35 | #73= | 47.3 | 68.2 | 74.0 | 87.4 | 61.0 |
| 72= | Blink v0.3 26B-A4B (NVFP4) | Full fine-tune | 25.8B | 55.9 | 57.2 | 42.1 | 58.2 | 4.0 | 25 | #74= | 45.1 | 63.6 | 63.0 | 70.0 | 47.0 |
| 72= | JADE | LoRA | 27.8B | 55.6 | 54.3 | 44.9 | 53.4 | 23.8 | 133 | #75= | 41.2 | 56.5 | 54.4 | 73.2 | 39.0 |
| 72= | Wald-26B-A4B | Full fine-tune | 25.81B | 55.4 | 56.4 | 41.9 | 62.3 | 4.3 | 90 | #76= | 44.7 | 64.3 | 69.1 | 83.3 | 50.9 |
| 72= | AJev lora1 (Gemma 4 26B-A4B) | LoRA | 25.81B | 55.3 | 54.9 | 43.6 | 57.7 | 4.8 | 89 | #77= | 41.9 | 62.1 | 63.5 | 73.2 | 46.8 |
| 77= | Cloudflare clef | Full fine-tune | 27.8B | 54.9 | 53.1 | 45.1 | 61.7 | 4.0 | 102 | #78= | 52.8 | 61.3 | 62.8 | 80.7 | 48.9 |
| 77= | StartLux-Decision v1.1 4B | Full fine-tune | 4.66B | 54.8 | 58.6 | 38.1 | 67.8 | 5.2 | 38 | #79= | 50.2 | 71.2 | 73.0 | 86.4 | 60.0 |
| 77= | Decider chat · Qwen3.6-27B | Inference technique | 27.78B | 54.7 | 52.2 | 45.7 | 51.6 | 2.1 | 84 | #80= | 38.3 | 55.9 | 52.2 | 71.4 | 37.0 |
| 77= | Perplexity Decider Flash v1.2 (4B) | Full fine-tune | 4.54B | 54.3 | 56.2 | 40.0 | 60.1 | 5.2 | 75 | #81= | 41.4 | 65.1 | 65.2 | 81.0 | 46.6 |
| 77= | vLLM-SR d3-flash (9B) | Full fine-tune | 8.39B | 54.3 | 54.6 | 41.9 | 57.8 | 5.6 | 54 | #82= | 43.3 | 60.6 | 64.2 | 76.0 | 41.9 |
| 77= | JPT-35B-A3B | LoRA | 35.95B | 54.1 | 52.3 | 44.4 | 52.7 | 4.9 | 250 | #83= | 36.5 | 61.2 | 55.6 | 69.2 | 36.6 |
| 83= | Xor 1.2 | LoRA | 35.95B | 53.6 | 53.6 | 41.8 | 52.3 | 2.4 | 150 | #84= | 35.5 | 56.9 | 54.5 | 71.2 | 45.0 |
| 83= | GEV-26B-Decide | LoRA | 25.8B | 53.5 | 55.0 | 39.9 | 58.0 | 7.2 | 129 | #85= | 43.7 | 61.2 | 67.9 | 69.7 | 45.6 |
| 83= | OneJev1.6-9B | Full fine-tune | 9.41B | 53.1 | 56.9 | 36.7 | 65.3 | 2.2 | 42 | #86= | 50.1 | 66.6 | 71.7 | 83.0 | 56.0 |
| 86= | Nace.AI Drex v2.0 (35B-A3B) | Full fine-tune | 35.11B | 52.8 | 54.3 | 39.2 | 61.3 | 1.8 | 276 | #87= | 47.9 | 65.3 | 63.0 | 76.5 | 53.9 |
| 86= | muria (Gemma 4 12B) | Full fine-tune | 11.96B | 52.6 | 53.5 | 39.8 | 51.4 | 7.1 | 54 | #88= | 31.4 | 56.6 | 58.9 | 70.8 | 39.1 |
| 86= | Jeffan-4B-2010 | Full fine-tune | 4.21B | 52.2 | 58.0 | 33.5 | 64.9 | 3.9 | 50 | #89= | 49.5 | 69.8 | 62.8 | 87.4 | 54.6 |
| 86= | Verdin-4B v1.1 | Full fine-tune | 4.21B | 52.0 | 54.1 | 37.9 | 61.0 | 6.3 | 200 | #90 | 43.7 | 63.3 | 66.9 | 81.0 | 51.1 |
| 90= | Bespoke Nimble 9B v3 | LoRA | 9.65B | 51.1 | 52.3 | 38.2 | 57.2 | 12.3 | 74 | #91= | 41.0 | 60.7 | 55.5 | 84.2 | 44.1 |
| 90= | Senova-4B v2 | Full fine-tune | 4.54B | 51.0 | 55.8 | 33.9 | 66.9 | 14.1 | 53 | #92= | 48.9 | 69.6 | 73.2 | 85.4 | 59.7 |
| 90= | frontier-infra Jebadiah 9B v2.1 (t1e) | LoRA | 9.65B | 50.9 | 51.0 | 39.5 | 51.5 | 5.5 | 125 | #93= | 35.3 | 56.6 | 55.5 | 70.4 | 38.1 |
| 90= | decisio 0.4.0 (Qwen3.6-35B-A3B) | Inference technique | 35.95B | 50.8 | 49.2 | 41.6 | 48.4 | 5.5 | 71 | #94= | 33.4 | 50.1 | 53.6 | 68.4 | 36.1 |
| 94= | vLLM-SR d3-mini (4B) | Full fine-tune | 4.54B | 50.0 | 51.9 | 36.5 | 54.9 | 5.6 | 50 | #95= | 39.8 | 57.8 | 63.6 | 72.0 | 37.7 |
| 94= | JoshuaSP diffusiongemma (open-jev) | Inference technique | 25.82B | 49.5 | 50.0 | 37.9 | 50.1 | 21.6 | 260 | #96= | 32.5 | 54.3 | 58.4 | 70.2 | 32.0 |
| 96= | AJev lora7 (Gemma 4 12B) | LoRA | 11.96B | 48.9 | 50.5 | 36.1 | 55.2 | 8.2 | 69 | #97= | 38.3 | 62.6 | 63.0 | 68.4 | 39.8 |
| 96= | OneJev1.6-4B | Full fine-tune | 4.54B | 48.9 | 55.8 | 29.6 | 62.0 | 2.5 | 27 | #98= | 45.0 | 64.4 | 68.9 | 81.5 | 49.8 |
| 96= | decisio 0.4.0 (Gemma 4 12B) | Inference technique | 11.96B | 48.8 | 48.4 | 38.3 | 49.7 | 10.3 | 38 | #99= | 30.8 | 52.4 | 58.0 | 71.6 | 35.3 |
| 99= | Mica v0.5 4B | Full fine-tune | 4.66B | 47.7 | 52.3 | 31.5 | 57.6 | 10.1 | 20 | #100= | 38.8 | 63.1 | 61.7 | 80.9 | 40.8 |
| 99= | PurpleMIST-Flash-1.0 | LoRA | 7.94B | 47.5 | 48.2 | 36.1 | 46.7 | 3.2 | 53 | #101= | 34.9 | 53.1 | 47.5 | 65.1 | 25.5 |
| 99= | Cerno 9B v1.1 | Full fine-tune | 8.95B | 47.3 | 52.4 | 30.6 | 58.6 | 4.8 | 27 | #102= | 39.3 | 64.3 | 65.1 | 79.2 | 43.0 |
| 99= | Sieve 9B v2 | LoRA | 9.65B | 47.0 | 49.3 | 33.7 | 53.4 | 7.3 | 51 | #103= | 38.3 | 55.9 | 64.2 | 69.8 | 34.0 |
| 103= | Decider 35B-A3B | Full fine-tune | 35.95B | 46.6 | 45.1 | 38.1 | 47.7 | 2.3 | 101 | #104= | 35.9 | 52.8 | 54.7 | 56.5 | 35.2 |
| 103= | Winnow-12B | LoRA | 11.96B | 46.6 | 48.1 | 34.3 | 50.7 | 16.8 | 73 | #105= | 34.8 | 57.0 | 54.1 | 71.0 | 31.4 |
| 103= | Seb-9B v2 | LoRA | 9.65B | 46.3 | 52.9 | 27.8 | 59.2 | 11.1 | 89 | #106= | 39.9 | 62.2 | 63.5 | 83.1 | 48.9 |
| 103= | vLLM-SR Decision 2.0 Lux 9B | Full fine-tune | 9.65B | 46.0 | 47.4 | 34.0 | 46.8 | 5.6 | 37 | #107= | 35.5 | 46.9 | 55.7 | 60.4 | 33.1 |
| 107= | Hopper 12B (trained) | LoRA | 11.96B | 45.8 | 47.0 | 34.1 | 48.3 | 7.6 | 290 | #109= | 31.7 | 53.8 | 53.2 | 66.3 | 34.2 |
| 107= | Nace.AI Drex v1.5 | Full fine-tune | 8.95B | 45.8 | 49.7 | 30.8 | 58.1 | 9.0 | 54 | #108= | 44.6 | 60.4 | 61.7 | 75.0 | 48.8 |
| 107= | Trio-Spark DI (9B) | Full fine-tune | 8.95B | 45.6 | 50.5 | 29.4 | 55.5 | 12.7 | 209 | #110= | 39.2 | 59.4 | 58.8 | 76.6 | 42.8 |
| 107= | CoCo-Decision-4B 2.4 | LoRA | 4.21B | 45.5 | 51.7 | 27.8 | 55.2 | 11.4 | 62 | #111= | 35.6 | 62.6 | 58.7 | 77.3 | 39.1 |
| 107= | Cloudflare clef-omni (30B-A3B) | Full fine-tune | 31.83B | 45.0 | 46.5 | 33.2 | 56.5 | 4.0 | 191 | #112= | 53.8 | 49.8 | 59.3 | 72.6 | 46.4 |
| 112= | JevK5 9B v0.3.3 | LoRA | 9.65B | 44.9 | 48.2 | 30.8 | 47.6 | 6.0 | 40 | #113= | 31.6 | 54.7 | 50.0 | 64.1 | 35.7 |
| 112= | Cloudflare clef-flash | Full fine-tune | 9.65B | 44.4 | 44.0 | 35.0 | 56.2 | 2.5 | 53 | #114= | 53.0 | 47.2 | 52.2 | 81.9 | 48.4 |
| 112= | Jevfire | Inference technique | 27.78B | 44.4 | 46.9 | 31.4 | 49.8 | 5.2 | 90 | #115= | 30.3 | 53.9 | 56.2 | 72.3 | 35.7 |
| 112= | RSI-Jev v6.1 VL 4B | Full fine-tune | 4.66B | 44.3 | 48.0 | 29.8 | 51.0 | 4.9 | 58 | #116= | 31.6 | 55.3 | 57.1 | 68.4 | 45.7 |
| 112= | Jegue 0.1 (9B) | LoRA | 9.65B | 44.2 | 48.8 | 28.6 | 54.1 | 11.6 | 46 | #117= | 33.5 | 62.5 | 65.5 | 72.3 | 29.8 |
| 112= | JPT-9B | LoRA | 9.65B | 44.1 | 43.5 | 35.0 | 47.0 | 7.7 | 149 | #119= | 30.9 | 57.7 | 44.6 | 67.0 | 28.8 |
| 112= | jiwo 4B v2 | Full fine-tune | 4.21B | 44.1 | 47.2 | 30.4 | 46.1 | 6.8 | 50 | #118= | 29.1 | 52.9 | 51.7 | 63.1 | 29.8 |
| 119= | ezjev 4B s2 | LoRA | 4.66B | 43.9 | 46.2 | 31.2 | 50.8 | 6.3 | 36 | #120= | 31.0 | 60.5 | 56.3 | 69.9 | 31.2 |
| 119= | Jev-Omni | LoRA | 11.96B | 43.8 | 42.2 | 36.0 | 41.0 | 16.1 | 55 | #121= | 26.1 | 51.3 | 35.8 | 60.8 | 27.4 |
| 119= | Solomon v1.1 | Head / adapter | 27.78B | 43.2 | 37.0 | 41.3 | 38.1 | 8.1 | 224 | #122= | 27.8 | 56.7 | 26.0 | 47.0 | 24.5 |
| 122= | Standard One 8B SH (v4) | LoRA | 8.92B | 42.9 | 50.4 | 24.1 | 57.4 | 10.7 | 90 | #123= | 36.9 | 57.9 | 66.4 | 84.0 | 42.5 |
| 122= | Amber Hybrid 4B | Inference technique | 4.61B | 42.7 | 50.4 | 23.7 | 59.9 | 4.6 | 274 | #124= | 36.7 | 60.1 | 68.9 | 84.9 | 55.3 |
| 122= | Saif 4B | LoRA | 4.21B | 42.5 | 44.2 | 30.8 | 50.8 | 5.0 | 95 | #125= | 29.9 | 54.4 | 62.0 | 71.3 | 35.1 |
| 122= | JEV-9B | Full fine-tune | 9.65B | 42.2 | 42.8 | 31.9 | 43.3 | 7.3 | 50 | #126= | 30.7 | 44.2 | 46.5 | 62.1 | 32.8 |
| 122= | Bespoke Nimble 9B v2 | LoRA | 9.65B | 42.1 | 39.5 | 35.7 | 39.9 | 2.4 | 77 | #127= | 30.5 | 44.3 | 34.9 | 59.2 | 27.6 |
| 127= | Sieve 9B | LoRA | 9.65B | 41.7 | 42.9 | 31.0 | 41.9 | 4.3 | 50 | #128= | 29.0 | 45.6 | 46.1 | 61.3 | 21.9 |
| 127= | Cerno 4B v1.1 | Full fine-tune | 4.21B | 41.2 | 47.4 | 24.2 | 57.6 | 5.0 | 27 | #129= | 34.5 | 63.0 | 65.9 | 82.9 | 40.6 |
| 127= | Sieve 4B | LoRA | 4.66B | 41.0 | 45.2 | 26.6 | 50.1 | 3.1 | 93 | #130= | 33.9 | 53.9 | 61.6 | 66.8 | 28.3 |
| 130= | Kev 9B v2 | LoRA | 9.65B | 40.9 | 42.4 | 29.9 | 41.5 | 4.5 | 60 | #131= | 30.8 | 46.7 | 46.8 | 56.9 | 17.5 |
| 130= | Nari Decision 4B v3.2 | Full fine-tune | 4.22B | 40.9 | 43.4 | 28.6 | 44.8 | 6.9 | 23 | #132= | 28.7 | 44.5 | 59.8 | 62.1 | 25.4 |
| 130= | Aplomb 1 | LoRA | 5.3B | 40.8 | 42.3 | 29.7 | 43.5 | 7.5 | 54 | #133= | 29.9 | 47.2 | 44.5 | 61.6 | 34.0 |
| 130= | vLLM-SR Decision 2.0 Nox 4B | LoRA | 4.21B | 40.8 | 43.7 | 27.9 | 44.2 | 3.0 | 34 | #134= | 31.4 | 46.4 | 52.4 | 60.1 | 26.3 |
| 130= | Hopper (G) 1.2 | LoRA | 4.66B | 40.2 | 42.4 | 28.4 | 41.2 | 9.3 | 23 | #135= | 24.6 | 45.7 | 45.4 | 60.1 | 29.2 |
| 130= | StartLux-Decision v1.1 2B | Full fine-tune | 2.27B | 40.1 | 51.5 | 17.0 | 59.2 | 3.5 | 30 | #136= | 36.2 | 62.4 | 69.7 | 79.0 | 53.7 |
| 136= | FlyMy.AI Decision 4B v1.6 | LoRA | 4.66B | 39.8 | 43.1 | 26.8 | 42.7 | 4.6 | 27 | #137= | 26.4 | 46.3 | 45.3 | 59.8 | 39.0 |
| 136= | Standard One 3B SH (v4) | LoRA | 3.85B | 39.8 | 49.8 | 18.5 | 58.1 | 11.4 | 80 | #138= | 36.3 | 57.5 | 67.0 | 85.7 | 47.5 |
| 136= | JPT-4B | LoRA | 4.66B | 39.6 | 40.1 | 29.9 | 42.8 | 7.9 | 139 | #139= | 26.5 | 53.7 | 45.0 | 57.2 | 26.1 |
| 136= | Amber 4B | LoRA | 4.21B | 39.5 | 48.2 | 19.8 | 52.1 | 5.7 | 128 | #140= | 34.7 | 58.3 | 54.6 | 71.9 | 40.1 |
| 140= | Intern-Decision-4B | Full fine-tune | 4.66B | 38.7 | 39.1 | 29.4 | 38.0 | 2.8 | 44 | #141= | 23.0 | 42.5 | 40.3 | 55.7 | 28.2 |
| 140= | RSI-Jev v5.0 VL 3B | Full fine-tune | 4.66B | 38.7 | 38.7 | 29.9 | 38.1 | 4.8 | 19 | #142= | 24.1 | 45.0 | 43.0 | 52.8 | 19.9 |
| 140= | djev | Inference technique | 25.82B | 38.2 | 41.6 | 25.2 | 40.9 | 21.2 | 84 | #143= | 25.9 | 50.0 | 41.8 | 51.7 | 34.4 |
| 143= | Valen-4B | Full fine-tune | 4.54B | 37.9 | 39.1 | 27.7 | 38.7 | 10.7 | 236 | #144= | 27.4 | 46.3 | 41.9 | 52.1 | 17.4 |
| 143= | razorback16 openjev diffusiongemma (NVFP4, vLLM) | Inference technique | 25.82B | 37.6 | 39.7 | 26.4 | 37.8 | 23.2 | 38 | #145= | 25.7 | 45.5 | 37.0 | 47.2 | 33.1 |
| 143= | NeoHorse-Jev-4B | Head / adapter | 4.66B | 37.3 | 36.9 | 29.2 | 37.0 | 10.4 | 49 | #146= | 23.6 | 39.6 | 43.9 | 58.9 | 11.0 |
| 146= | JevK5 v0.3 | LoRA | 4.66B | 36.6 | 41.8 | 21.9 | 41.1 | 6.1 | 26 | #147= | 25.9 | 46.6 | 44.5 | 56.9 | 30.8 |
| 146= | Decider 4B | Full fine-tune | 4.66B | 36.3 | 37.2 | 26.9 | 41.2 | 8.4 | 13 | #148= | 27.7 | 44.0 | 44.7 | 58.6 | 30.0 |
| 146= | baselabs joey-1 (2B) | Full fine-tune | 2.72B | 35.9 | 50.2 | 10.2 | 63.3 | 4.4 | 85 | #149= | 45.1 | 62.8 | 71.2 | 85.8 | 54.7 |
| 149= | lev | LoRA | 4.66B | 35.5 | 37.2 | 25.2 | 39.2 | 6.4 | 71 | #150= | 22.8 | 39.1 | 47.9 | 57.5 | 30.8 |
| 149= | Lev1 3.7B | LoRA | 3.7B | 35.4 | 41.5 | 19.9 | 42.7 | 5.8 | 29 | #151= | 23.1 | 44.3 | 47.1 | 68.8 | 32.4 |
| 149= | Kev 4B r10 | LoRA | 4.66B | 34.9 | 36.2 | 25.4 | 38.0 | 3.2 | 59 | #152= | 26.3 | 41.8 | 42.5 | 55.0 | 17.9 |
| 152= | Jev-Style 2B v4 | Full fine-tune | 2.21B | 34.5 | 41.9 | 17.6 | 44.4 | 11.0 | 40 | #153= | 23.6 | 42.5 | 65.4 | 61.8 | 29.0 |
| 152= | Sieve 2B v3 | LoRA | 2.27B | 34.5 | 45.8 | 12.6 | 53.5 | 7.2 | 43 | #154= | 30.9 | 49.3 | 66.5 | 80.5 | 47.6 |
| 152= | Cam1 | LoRA | 4.66B | 34.4 | 32.3 | 29.0 | 30.7 | 6.3 | 249 | #157= | 19.2 | 33.8 | 36.9 | 46.3 | 11.5 |
| 152= | LatentMesh Blitz 2B v1 | LoRA | 2.27B | 34.4 | 44.2 | 14.5 | 48.2 | 13.7 | 52 | #155= | 22.3 | 54.0 | 60.4 | 66.5 | 41.8 |
| 152= | Saracura PT-BR 4B | LoRA | 4.66B | 34.4 | 35.0 | 25.9 | 38.0 | 17.9 | 62 | #156= | 25.3 | 41.9 | 46.2 | 53.9 | 14.9 |
| 152= | mmastrac diffusiongemma (vLLM PR 57250) | Inference technique | 25.82B | 34.3 | 33.4 | 27.4 | 32.7 | 20.2 | 127 | #159= | 25.0 | 46.5 | 24.2 | 34.0 | 31.7 |
| 152= | vLLM-SR d3-nano (2B) | Full fine-tune | 2.21B | 34.3 | 42.5 | 16.4 | 42.2 | 11.2 | 38 | #158= | 25.3 | 42.2 | 58.8 | 58.9 | 22.4 |
| 152= | Kas 4B | LoRA | 4.02B | 34.0 | 38.7 | 20.4 | 38.4 | 9.8 | 69 | #160= | 32.2 | 35.1 | 40.4 | 48.9 | 39.4 |
| 152= | Tev1-4B-experimental | Full fine-tune | 4.66B | 33.7 | 31.8 | 28.4 | 29.5 | 10.4 | 36 | #161= | 24.3 | 39.2 | 15.5 | 41.8 | 23.5 |
| 161= | Winnow-E4B | LoRA | 8B | 32.7 | 36.8 | 20.2 | 40.9 | 5.8 | 45 | #162= | 22.8 | 46.8 | 43.8 | 62.5 | 26.9 |
| 161= | Aurelis-2B | Full fine-tune | 2.52B | 32.6 | 44.7 | 10.1 | 51.3 | 6.5 | 74 | #163= | 32.2 | 54.1 | 54.2 | 79.5 | 35.9 |
| 161= | Jevstral 8B | LoRA | 8B | 32.2 | 36.3 | 19.8 | 34.0 | 4.7 | 28 | #164= | 24.7 | 33.6 | 39.2 | 52.9 | 14.1 |
| 161= | Metask-Jev-4B | LoRA | 4.66B | 32.2 | 27.8 | 30.1 | 27.6 | 10.4 | 57 | #165= | 21.6 | 40.1 | 12.9 | 37.6 | 21.9 |
| 161= | Ryotide Qwen v0.3.1 | Inference technique | 4B | 32.0 | 36.7 | 18.8 | 37.3 | 9.2 | 81 | #166= | 21.0 | 36.0 | 47.5 | 55.1 | 29.8 |
| 166= | Intelif Qwen3-4B | LoRA | 4.02B | 31.3 | 35.9 | 18.4 | 32.5 | 7.1 | 23 | #167= | 19.5 | 31.2 | 39.6 | 51.0 | 21.3 |
| 166= | GLiClass v4 edge (0.4B) | Full fine-tune | 0.44B | 31.2 | 42.2 | 10.5 | 47.5 | 8.9 | 22 | #168= | 21.4 | 54.6 | 66.7 | 58.1 | 38.6 |
| 166= | LiquidAI d1-3B | Full fine-tune | 3.12B | 31.1 | 41.2 | 11.5 | 49.0 | 4.7 | 16 | #169= | 24.3 | 56.5 | 52.8 | 74.5 | 38.9 |
| 166= | open-jev (pngwn) | LoRA | 4.66B | 31.1 | 30.0 | 25.2 | 30.5 | 5.9 | 112 | #170= | 22.8 | 34.7 | 29.1 | 45.7 | 14.5 |
| 166= | vera-core-2 | Full fine-tune | 1.88B | 30.6 | 39.6 | 12.5 | 41.3 | 4.6 | 90 | #171= | 21.4 | 43.9 | 53.8 | 61.5 | 23.9 |
| 171= | Jobe | Inference technique | 4.66B | 30.4 | 34.3 | 18.4 | 32.9 | 12.3 | 54 | #172= | 18.5 | 34.8 | 36.7 | 48.8 | 28.9 |
| 171= | OneJev1.6-0.8B | Full fine-tune | 0.85B | 29.6 | 40.5 | 9.3 | 48.8 | 3.5 | 13 | #173= | 26.1 | 48.1 | 63.4 | 72.9 | 35.6 |
| 173= | crystal-0.8b | Full fine-tune | 0.85B | 29.4 | 39.3 | 10.3 | 42.8 | 1.7 | 54 | #174= | 19.1 | 44.8 | 55.4 | 64.9 | 33.4 |
| 173= | Crivo-1B | Full fine-tune | 1.03B | 29.3 | 38.8 | 10.8 | 47.8 | 15.1 | 16 | #176= | 18.5 | 56.3 | 55.8 | 78.7 | 28.9 |
| 173= | vLLM-SR vela3-fast (Vela 3.0, 0.4B) | Full fine-tune | 0.38B | 29.3 | 40.2 | 9.0 | 50.4 | 12.1 | 54 | #175= | 21.2 | 53.1 | 66.9 | 73.7 | 43.1 |
| 173= | Gevva E4B | Full fine-tune | 8B | 29.2 | 31.0 | 20.2 | 32.4 | 14.5 | 53 | #177= | 17.9 | 27.6 | 40.0 | 56.9 | 22.6 |
| 173= | StartLux-Decision v1.1 0.8B | Full fine-tune | 0.87B | 28.9 | 41.3 | 6.9 | 53.8 | 1.5 | 30 | #178= | 28.5 | 55.5 | 66.5 | 74.6 | 51.1 |
| 173= | openvons | Inference technique | 4.02B | 28.8 | 32.1 | 17.9 | 28.7 | 37.1 | 21 | #179= | 19.5 | 30.7 | 25.4 | 44.7 | 24.4 |
| 179= | SemIf | Inference technique | 4.66B | 26.7 | 28.0 | 18.7 | 26.5 | 9.9 | 113 | #180= | 17.9 | 34.0 | 15.7 | 39.1 | 27.9 |
| 179= | Saina Helm 2 Omni (0.8B) | LoRA | 0.85B | 26.5 | 35.2 | 9.6 | 37.4 | 10.0 | 39 | #181= | 15.3 | 39.3 | 50.2 | 59.7 | 23.7 |
| 179= | vLLM-SR d3-lite (0.8B) | Full fine-tune | 0.85B | 26.2 | 34.8 | 9.3 | 36.9 | 9.4 | 37 | #182= | 18.6 | 34.2 | 53.7 | 56.3 | 22.3 |
| 179= | Amber 0.4B | Full fine-tune | 0.4B | 26.0 | 35.7 | 7.9 | 49.7 | 7.5 | 43 | #183= | 18.9 | 48.7 | 64.4 | 78.9 | 49.6 |
| 183= | jiwo 0.8B v2 | Full fine-tune | 0.75B | 25.8 | 30.7 | 13.7 | 29.6 | 6.8 | 36 | #184= | 10.9 | 36.1 | 44.2 | 40.3 | 12.2 |
| 183= | Saina Helm 2.0 v2 | LoRA | 0.85B | 25.3 | 34.7 | 7.8 | 39.3 | 16.8 | 39 | #185= | 16.6 | 44.1 | 50.5 | 61.3 | 22.5 |
| 185= | this-that 1.2 | Full fine-tune | 1.88B | 24.9 | 27.5 | 15.7 | 28.7 | 19.8 | 44 | #186= | 16.6 | 31.5 | 32.8 | 45.0 | 14.2 |
| 185= | Gevva E2B | Full fine-tune | 5.1B | 24.3 | 27.8 | 14.1 | 28.5 | 9.0 | 43 | #187= | 11.9 | 25.0 | 38.6 | 52.5 | 16.5 |
| 187= | Crivo-400M | Full fine-tune | 0.4B | 23.7 | 30.7 | 9.5 | 39.8 | 12.1 | 17 | #188= | 11.5 | 44.3 | 51.0 | 69.0 | 25.4 |
| 187= | Decider 2B | Full fine-tune | 2.27B | 23.7 | 28.5 | 12.0 | 29.8 | 7.7 | 8 | #189= | 17.2 | 32.2 | 37.3 | 42.4 | 17.6 |
| 187= | jdi-ptr14-0.8b | Full fine-tune | 0.75B | 23.6 | 31.3 | 8.4 | 35.8 | 4.2 | 67 | #191= | 13.5 | 40.3 | 52.6 | 53.0 | 16.8 |
| 187= | vLLM-SR Decision 2.0 Sol 2B | Full fine-tune | 2.27B | 23.6 | 29.3 | 11.1 | 30.1 | 3.7 | 26 | #190= | 16.2 | 24.6 | 45.6 | 48.8 | 14.4 |
| 191= | mini-jev | Inference technique | 4.02B | 22.8 | 23.4 | 16.6 | 21.3 | 34.6 | 67 | #192= | 15.8 | 29.8 | 8.7 | 28.4 | 25.3 |
| 191= | Valen-2B | Full fine-tune | 2.21B | 22.2 | 24.3 | 14.2 | 27.0 | 10.5 | 161 | #193= | 14.8 | 30.4 | 26.2 | 45.1 | 17.8 |
| 191= | vLLM-SR vela3-fast-vision (Vela 3.0, 0.25B) | Full fine-tune | 0.25B | 22.1 | 30.5 | 6.5 | 37.5 | 11.2 | 37 | #194= | 13.2 | 36.6 | 54.9 | 51.8 | 41.5 |
| 194= | Sieve 2B | LoRA | 2.27B | 21.9 | 23.7 | 14.4 | 22.2 | 8.9 | 63 | #195= | 10.4 | 20.8 | 28.5 | 35.3 | 19.5 |
| 194= | wyat-1-800m | LoRA | 0.87B | 21.5 | 25.2 | 11.7 | 25.8 | 23.2 | 112 | #196= | 11.3 | 20.4 | 43.0 | 41.9 | 13.2 |
| 196= | Intern-Decision-2B | Full fine-tune | 2.27B | 20.9 | 23.3 | 12.9 | 19.3 | 6.1 | 34 | #197= | 12.0 | 21.3 | 15.7 | 33.3 | 14.4 |
| 196= | Sieve 0.8B | Full fine-tune | 0.59B | 20.8 | 26.0 | 9.3 | 32.3 | 8.8 | 52 | #198= | 14.3 | 26.6 | 54.8 | 51.2 | 13.5 |
| 196= | Strands Decider 2B | LoRA | 1.9B | 20.7 | 22.4 | 13.5 | 25.4 | 4.4 | 52 | #199= | 11.2 | 25.7 | 34.0 | 43.8 | 10.7 |
| 196= | EXAONE-4.0-1.2B-JEV v0.3 | Full fine-tune | 1.28B | 20.6 | 26.9 | 7.8 | 30.7 | 7.9 | 149 | #200= | 15.0 | 26.6 | 40.9 | 56.6 | 13.9 |
| 200= | tasksource Decider nano v5 (150M) | Full fine-tune | 0.15B | 19.9 | 26.1 | 7.3 | 33.5 | 7.0 | 13 | #201= | 9.7 | 38.7 | 50.1 | 50.9 | 16.6 |
| 200= | Crivo-68M | Full fine-tune | 0.07B | 19.5 | 25.3 | 7.6 | 32.8 | 10.3 | 12 | #202= | 5.0 | 30.8 | 48.8 | 64.0 | 20.9 |
| 202= | JEV Ettin 150M | Full fine-tune | 0.15B | 18.5 | 25.5 | 5.4 | 37.3 | 12.9 | 77 | #203= | 14.5 | 43.2 | 52.1 | 60.4 | 9.4 |
| 202= | vLLM-SR Decision 2.0 Eos 0.8B | Full fine-tune | 0.87B | 18.4 | 21.3 | 10.2 | 20.6 | 13.0 | 25 | #204= | 7.0 | 15.3 | 38.4 | 34.0 | 8.9 |
| 202= | Crivo-150M v2 | Full fine-tune | 0.15B | 18.3 | 24.9 | 5.5 | 34.4 | 11.5 | 13 | #205= | 8.3 | 34.5 | 48.2 | 62.8 | 21.8 |
| 202= | Bekko System One v0 400M | Full fine-tune | 0.4B | 18.0 | 21.6 | 9.1 | 22.5 | 11.1 | 17 | #206= | 11.6 | 24.0 | 42.2 | 18.6 | 13.9 |
| 202= | Mola-150M | Full fine-tune | 0.15B | 17.8 | 23.6 | 6.3 | 31.3 | 8.0 | 13 | #207= | 12.0 | 30.2 | 51.4 | 50.1 | 9.9 |
| 207= | vLLM-SR d3-edge (0.6B) | Full fine-tune | 0.59B | 17.7 | 24.3 | 5.0 | 28.8 | 14.1 | 19 | #208= | 11.1 | 24.2 | 53.6 | 42.7 | 11.8 |
| 207= | Mola-68M | Full fine-tune | 0.07B | 17.6 | 22.0 | 7.7 | 27.3 | 11.9 | 11 | #209= | 7.9 | 23.3 | 49.4 | 47.9 | 5.7 |
| 207= | JEV FLAN-T5 Large public v2 (342M) | Full fine-tune | — | 17.3 | 19.8 | 9.9 | 22.0 | 4.7 | 17 | #210= | 12.4 | 29.7 | 35.5 | 17.6 | 8.1 |
| 207= | DecisionTune 1.0 | Full fine-tune | 0.39B | 17.2 | 22.1 | 6.8 | 28.9 | 14.8 | 14 | #211= | 11.7 | 29.9 | 45.0 | 46.5 | 5.7 |
| 211= | Valen-0.8B | Full fine-tune | 0.85B | 16.6 | 19.6 | 8.6 | 20.8 | 10.8 | 200 | #212= | 10.2 | 24.8 | 20.6 | 35.6 | 11.2 |
| 211= | Bosun v3.1 1.7B | LoRA | 1.72B | 16.5 | 20.4 | 7.5 | 20.6 | 13.0 | 40 | #213= | 4.8 | 13.5 | 42.8 | 34.8 | 9.0 |
| 211= | Dinah-1 | Full fine-tune | 0.15B | 16.4 | 22.3 | 4.9 | 34.3 | 21.8 | 14 | #215= | 23.1 | 23.1 | 50.7 | 62.2 | 8.6 |
| 211= | vLLM-SR Decision 2.0 Kai 0.6B | Full fine-tune | 0.6B | 16.4 | 16.9 | 11.6 | 16.7 | 3.8 | 19 | #214= | 6.4 | 11.6 | 32.0 | 26.7 | 7.4 |
| 211= | Kev 0.8B r15 | LoRA | 0.87B | 16.3 | 15.8 | 12.7 | 16.8 | 4.0 | 45 | #216= | 7.0 | 14.2 | 22.1 | 35.5 | 3.8 |
| 216= | PICK-150M v5 | Full fine-tune | 0.15B | 15.6 | 19.8 | 6.5 | 32.1 | 12.5 | 16 | #217= | 8.0 | 35.5 | 45.2 | 59.9 | 8.4 |
| 216= | JPT-0.8B | LoRA | 0.87B | 15.5 | 17.8 | 8.6 | 19.4 | 6.9 | 118 | #218= | 8.2 | 25.2 | 16.4 | 36.2 | 8.4 |
| 216= | JEV Ettin 400M | Full fine-tune | 0.4B | 15.2 | 14.8 | 11.8 | 14.1 | 9.0 | 16 | #219= | 8.7 | 12.9 | 26.3 | 14.8 | 5.8 |
| 219= | Sifr 0.8B v3.1 | Full fine-tune | 0.87B | 14.5 | 18.9 | 5.3 | 25.7 | 12.0 | 35 | #220= | 13.2 | 23.1 | 45.5 | 35.9 | 5.8 |
| 219= | Crivo-32M v2 | Full fine-tune | 0.03B | 14.2 | 18.4 | 5.3 | 27.3 | 13.3 | 8 | #221= | 2.7 | 25.0 | 45.6 | 50.0 | 19.0 |
| 219= | Prumo | Inference technique | — | 14.0 | 15.8 | 8.1 | 20.1 | 4.7 | 0 | #222= | 17.0 | 3.1 | 26.6 | 45.3 | 13.4 |
| 219= | Kodiak v0.2 1B (accuracy) | Full fine-tune | 3B | 13.9 | 16.3 | 7.1 | 19.1 | 7.7 | 142 | #223= | 13.5 | 15.3 | 36.2 | 15.1 | 15.9 |
| 223= | Intern-Decision-0.8B | Full fine-tune | 0.87B | 12.7 | 12.9 | 9.0 | 11.8 | 2.6 | 34 | #224= | 6.8 | 10.9 | 6.4 | 28.4 | 7.7 |
| 223= | Jev-Style 0.8B v3 | Full fine-tune | 0.87B | 12.5 | 14.7 | 6.4 | 17.0 | 8.4 | 36 | #225= | 5.4 | 11.9 | 26.2 | 35.4 | 8.3 |
| 223= | Tev1-0.8B-experimental | Full fine-tune | 0.87B | 12.5 | 12.8 | 8.6 | 13.1 | 12.6 | 29 | #226= | 6.8 | 15.3 | 4.9 | 32.0 | 5.9 |
| 223= | Bekko System One v0 68M | Full fine-tune | 0.07B | 12.4 | 12.6 | 8.7 | 13.0 | 11.3 | 12 | #227= | 4.0 | 6.4 | 32.2 | 14.1 | 13.1 |
| 227= | LiquidAI d1-omni-600M | Full fine-tune | 0.59B | 11.8 | 12.9 | 7.1 | 17.9 | 12.8 | 11 | #228= | 6.6 | 19.3 | 35.0 | 19.5 | 6.5 |
| 227= | Crivo-17M | Full fine-tune | 0.02B | 11.7 | 15.9 | 3.2 | 22.2 | 8.0 | 6 | #229= | 1.2 | 21.0 | 40.1 | 36.5 | 17.0 |
| 227= | Von 1.3 | Full fine-tune | 0.4B | 11.7 | 12.8 | 7.0 | 13.9 | 21.6 | 18 | #230= | 9.3 | 11.6 | 30.1 | 11.7 | 2.7 |
| 230= | Bosun v3.1 0.6B | LoRA | 0.6B | 11.0 | 14.1 | 4.2 | 14.5 | 14.2 | 39 | #231= | 2.6 | 4.3 | 34.2 | 28.4 | 6.7 |
| 230= | vera-spark | Full fine-tune | 0.64B | 11.0 | 15.1 | 2.8 | 19.6 | 15.7 | 80 | #232= | 10.1 | 24.9 | 30.1 | 18.1 | 12.4 |
| 230= | Lavoir | Full fine-tune | 0.4B | 10.9 | 9.9 | 8.9 | 8.9 | 14.9 | 20 | #233= | 2.8 | 11.5 | 12.0 | 13.0 | 3.9 |
| 233 | GLiNER2.5-Decide | Full fine-tune | 0.49B | 10.1 | 12.1 | 4.9 | 11.5 | 8.8 | 23 | #234 | 2.3 | 8.3 | 25.3 | 14.8 | 9.8 |
| 234= | Laya | Full fine-tune | 0.42B | 9.2 | 9.1 | 6.6 | 6.2 | 14.0 | 6 | #235= | 2.9 | 10.1 | 7.2 | 5.8 | 3.5 |
| 234= | JevK5 Lite | Full fine-tune | 0.3B | 8.8 | 10.4 | 4.2 | 10.3 | 14.9 | 19 | #236= | 5.4 | 9.0 | 28.7 | 3.1 | 2.6 |
| 236= | LFM2.5-2.6B-RLCD | Full fine-tune | 2.7B | 8.2 | 7.7 | 6.3 | 6.8 | 25.5 | 39 | #237= | 6.6 | 5.9 | 10.5 | 4.4 | 6.3 |
| 236= | Qwen-2.5-1B-RLCD | Inference technique | 1.54B | 7.9 | 6.2 | 7.4 | 4.0 | 20.5 | 44 | #238= | 1.4 | 5.2 | 0.9 | 7.0 | 8.5 |
| 236= | Bekko System One v0 17M | Full fine-tune | 0.02B | 7.7 | 5.9 | 7.4 | 7.4 | 16.1 | 7 | #239= | 2.5 | 2.7 | 21.3 | 5.9 | 7.5 |
| 239= | CLM-v0.1-8B | Full fine-tune | 8.19B | 6.5 | 6.0 | 4.9 | 7.4 | 32.3 | 47 | #240= | 2.4 | 5.7 | 11.3 | 14.0 | 5.2 |
| 239= | jeff | Inference technique | 0.58B | 6.2 | 6.9 | 3.3 | 8.3 | 9.7 | 22 | #241= | 2.7 | 11.6 | 21.3 | 1.2 | 0.9 |
| 239= | LFM2.5-350M-RLCD | Full fine-tune | 0.35B | 5.9 | 2.3 | 8.2 | 1.4 | 56.8 | 27 | #242= | 0.8 | 0.6 | 0.0 | 3.8 | 3.6 |
| 242= | system-one-gemma | LoRA | 0.27B | 5.2 | 5.0 | 3.5 | 5.2 | 23.9 | 32 | #243= | 1.3 | 1.2 | 17.2 | 3.3 | 4.9 |
| 242= | DM-JEPA 1.1 | Full fine-tune | 0.15B | 5.0 | 4.2 | 4.0 | 19.2 | 44.5 | 63 | #244= | 17.3 | 21.7 | 34.0 | 7.9 | 8.6 |
| 242= | Julia 1 | Full fine-tune | 0.14B | 4.8 | 4.5 | 3.2 | 4.2 | 42.0 | 6 | #245= | 1.4 | 1.4 | 9.9 | 7.2 | 2.1 |
| 245= | Nirnaya 0.4B | Full fine-tune | 0.39B | 3.5 | 2.7 | 3.0 | 1.6 | 4.6 | 14 | #246= | 0.5 | 0.5 | 1.4 | 5.5 | 0.5 |
| 245= | Zev | Inference technique | — | 3.4 | 1.8 | 3.8 | 1.6 | 13.8 | 1 | #247= | 0.7 | 0.2 | 0.9 | 3.0 | 6.0 |
| 247 | Verdict | Full fine-tune | 0.15B | 2.0 | 2.4 | 0.3 | 1.9 | 15.4 | 11 | #248 | 0.3 | 1.6 | 0.0 | 6.6 | 1.5 |
Methodology
This leaderboard mirrors Decision Index 0.4.1. S1 Hub does not re-run or adjust any score — it imports the published file once a day and keeps a history of every import. Source: Decision Index on Hugging Face
Published by the source
Every number is the Decision Index’s own result on shared hardware; method details are on the source page.
Synced daily
A scheduled job imports the latest file every day at 03:00 UTC. Unchanged files are skipped; admins can see every run.
Calibration error
A decision model should know how sure it is. Lower ECE is better; shown separately, never scored.