# JevBench, modèles à poids ouverts, édition du 9 novembre 2026

Tableau repris tel que JevBench le publie : même score de capacité, pour des modèles à poids ouverts que le banc fait tourner lui-même sur une même carte graphique ; même règle de rang. Rang publié par le banc, jamais recalculé ; aucun score de QUELLE IA. Édition du 9 novembre 2026, version v1.8.7 du banc.

| Rang | Système | Capacité |
| ---: | --- | --- |
| 1 | baselabs mammoth-1 (27B) | 85,2 |
| 2 | KnowLine 31B-Gen1 | 85,2 |
| 3 | Fastino GLiDE 1.2 (open weights, self-hosted) | 84,9 |
| 4 | H2O-Lightning-31B v1.0.7 (H2O.ai, Gemma-4-31B-it fine-tune, vLLM + h2oai overlay, one-engine FP8 profile, self-hosted) | 84,3 |
| 5 | Perplexity Decider v1.1 27B (Qwen3.8-27B, noncausal decision readout) | 82,7 |
| 6 | vLLM-SR d3-turbo | 82,6 |
| 7 | StartLux-Decision v1.1 27B Instant (StartLux Labs, Qwen3.8-27B fine-tune) | 82,6 |
| 8 | vLLM-SR d3 | 82,1 |
| 9 | basal-omni-1.0-ultra (Remek Kinas, Gemma-4-31B-it + merged LoRA, typed-decision readout, self-hosted) | 82,1 |
| 10 | DOPEST 1.0 31B | 82,0 |
| 11 | Jeff-1.1-Medium (jgeuter) | 81,9 |
| 12 | Arcbright Elf 1 31B | 81,8 |
| 13 | Quyet-1.0-Large (Chinh Nguyen, Gemma-4-31B decoder, option-letter logits) | 81,7 |
| 14 | Jeffan-31b-1710 | 81,7 |
| 15 | Jeff-1.0-Large | 81,5 |
| 16 | Nimble-v4-max (open weights, self-hosted) | 81,2 |
| 17 | Sieve-31B-v2 | 81,1 |
| 18 | Turion Jev 31B | 80,9 |
| 19 | pplx-decider-v1.2 (Perplexity, Qwen3.8-27B full fine-tune, marker readout, self-hosted) | 80,8 |
| 20 | torchcast-decision-27b (Torchcast AI, Qwen3.8-27B fine-tune) | 80,2 |
| 21 | Lychee-27B | 79,8 |
| 22 | metask-jev-rain-31B-trained service v3 | 79,6 |
| 23 | decisio v0.8.0 on gemma-4-31B-it (frozen, FP8 on load, one prefill per state, self-hosted) | 79,6 |
| 24 | Vera Large | 79,6 |
| 25 | Kev 27B (Kev 1.0, Qwen3.8-27B) | 78,9 |
| 26 | metask-jev-rain-31B-trained | 78,9 |
| 27 | StartLux-Decision v1.1 27B AutoThink (StartLux Labs, Qwen3.8-27B fine-tune, gate 0.6) | 78,8 |
| 28 | Bobcat 1.3 (Gemma 4 31B) | 77,9 |
| 29 | deck-31B (krishna765, frozen Gemma-4-31B-it, TorchAO FP8 dynamic) | 77,6 |
| 30 | Torchcast Decision 4B (v1.3) | 77,5 |
| 31 | KnowLine 26B-A4B-Gen2.5 | 77,3 |
| 32 | Eikos-27B (caiovicentino1, Qwen3.8-27B) | 77,2 |
| 33 | metask-jev-rain-12B-trained | 76,9 |
| 34 | Wald 26B-A4B (Gemma 4 26B-A4B) | 76,8 |
| 35 | KnowLine 35B-A3B-Gen1.5 | 76,7 |
| 36 | René-1 31B FP8 (salfatigroup, Gemma 4 31B full fine-tune, one-pass option readout) | 76,0 |
| 37 | Decisio 12B tuned commit (variant of the Decisio family) | 75,7 |
| 38 | Clef (Cloudflare, Qwen3.8-27B post-train with a joint schema head, multimodal, measured on text) | 75,3 |
| 39 | H2O-Lightning-4B v1.1 (H2O.ai, Qwen3.5-4B fine-tune, stock vLLM + open shim) | 75,0 |
| 40 | JEV-27B (AutoTrust, Qwen3.8-27B) | 73,9 |
| 41 | Nagi-ENORMOUS 27B (nagisanzenin, Qwen3.8-27B LoRA) | 73,8 |
| 42 | Surogate Rune 26B-A4B v3 (v1.6 pool) | 73,5 |
| 43 | RSI-Jev v6.1-VL 27B | 73,5 |
| 44 | Bobcat Flash 1.2 (Gemma-4-26B-A4B-it + two merged rank-64 LoRA adapters, typed-decision readout at the first answer position, FP8, self-hosted) | 73,0 |
| 45 | Torchcast Decision 4B (v1.2) | 72,6 |
| 46 | swanOne (blockbrain, Qwen3.8-Flash-Next NVFP4) | 72,4 |
| 47 | FrontiersMind Lumma-fev-27B | 72,3 |
| 48 | torchcast-decision-12b (Torchcast AI, Gemma-4-12B fine-tune, option-letter logprob readout) | 71,7 |
| 49 | AutoJev-27B (denis-pplx, Qwen3.8-27B) | 71,6 |
| 50 | decider-12b v2 (Mapika) | 71,3 |
| 51 | Decision 2.0 Vega 27B (vLLM-SR, Qwen3.8-27B) | 71,3 |
| 52 | Jev-Omni (akhilaaa3, Gemma-4-12B merged) | 71,3 |
| 53 | Winnow-12B Q8 | 71,2 |
| 54 | Xor 26B-A4B (Juspay, Gemma-4-26B-A4B, bf16) | 71,2 |
| 55 | decider-12b v1, stock Gemma-4-12B-it (Mapika) | 71,0 |
| 56 | Cygnet (blockbrain, frozen Gemma-4-12B-it) | 70,9 |
| 57 | decisio v0.8.0 on gemma-4-12B-it (frozen, one forward pass, self-hosted) | 70,7 |
| 58 | JPT-35B-A3B (Qwen3.5-35B-A3B fine-tune) | 70,7 |
| 59 | decider chat on Gemma-4-31B-it (Mapika, frozen base, inference technique) | 70,4 |
| 60 | baselabs elephant-1 (9B) | 70,2 |
| 61 | Jebadiah 27B (Frontier Infra, Qwen3.8-27B LoRA) | 70,2 |
| 62 | GEV-26B-Decide (AutoTrust, Gemma-4-26B-A4B + LoRA + head) | 70,1 |
| 63 | Chon-1.0-12B-Fast | 70,1 |
| 64 | decisio v0.9.0 on gemma-4-12B-it (frozen, one prefill per question, self-hosted) | 70,0 |
| 65 | Blink v0.3 26B-A4B | 69,9 |
| 66 | NInfer Qwen3.8-Flash-Next mixed | 69,7 |
| 67 | Respan Span-02 | 69,3 |
| 68 | Philly Decider v2 (PhillyLabs) | 69,3 |
| 69 | Wald 4B v2.1 | 69,3 |
| 70 | JevOne | 68,9 |
| 71 | vLLM-SR d3-mini | 68,8 |
| 72 | Torchcast Decision 4B (v1.1) | 68,7 |
| 73 | iota-decide-4b-v0.3 (Macrocosmos, letter-readout 4B, self-hosted) | 68,6 |
| 74 | Xevn-4B | 68,6 |
| 75 | Chon-1.0-E4B | 68,5 |
| 76 | Hopper 12B trained (gemma-4-12B-it frozen + LoRA r32 unmerged, one forward pass, self-hosted) | 68,1 |
| 77 | SPX-CD-Omni (SurdAI, google/gemma-4-12B-it + LoRA r32, one forward pass, self-hosted) | 68,1 |
| 78 | NInfer Qwen3.8-27B NVFP4 | 68,0 |
| 79 | Aplomb 1 (5.3B decision model on Qwen3.5-4B, trained readout head, self-hosted) | 67,1 |
| 80 | Bespoke Nimble 9B v3 (Qwen3.5-9B LoRA) | 66,8 |
| 81 | Amazon Strands Decider 26B-A4B | 66,7 |
| 82 | Tern-35B-A3B (matthewagi, Qwen3.6 35B-A3B MoE decision model, author server, self-hosted) | 66,4 |
| 83 | KnowLine 4B-Gen5.5 | 66,2 |
| 84 | decider-35b-a3b (Mapika) | 66,2 |
| 85 | Metask rain 4B | 66,0 |
| 86 | shisa-de-1 (shisa-ai, Gemma-4-26B-A4B) | 65,5 |
| 87 | Plumb-4B (crh225, JevK5 v0.2 + LoRA) | 65,3 |
| 88 | Arcbright Elf 1 12B | 65,0 |
| 89 | decider-4b v2 (Mapika) | 64,9 |
| 90 | Seb-9B-v2 (ironbcc, Qwen3.5-9B fine-tune, one forward pass, stock vLLM, self-hosted) | 64,8 |
| 91 | JevK5 v0.3 (4B) | 64,2 |
| 92 | Jeffan-4B-2010 | 64,0 |
| 93 | Clef-Flash (Cloudflare, Qwen3.5-9B post-train with a joint schema head, multimodal, measured on text) | 63,8 |
| 94 | JPT-4B (kirp, Qwen3.5-4B LoRA) | 63,8 |
| 95 | deck-4B v1.0 (krishna765, JevK5 + LoRA, FP8 weight-only) | 63,6 |
| 96 | Jebadiah 9B v2.1 (Frontier Infra, Qwen3.5-9B LoRA) | 63,4 |
| 97 | GLiClass v4 9B (Knowledgator) | 63,2 |
| 98 | Seb-9B (ironbcc, Qwen3.5-9B fine-tune, one forward pass, stock vLLM, self-hosted) | 63,1 |
| 99 | janus 4B (Icarus AI / cmxu, Qwen3.5-4B + LoRA r64 and pointer decision head) | 62,9 |
| 100 | JevK5 v0.2.0 | 62,8 |
| 101 | Messier One v0.4 (Qwen3.5-4B fine-tune, self-hosted) | 62,4 |
| 102 | Imajev-9B (mohit67890, Qwen3.5-9B LoRA) | 62,4 |
| 103 | Imajev-4B (RTX 5090) | 61,9 |
| 104 | jqv (Qwen3-32B zero-shot) | 61,9 |
| 105 | metask-jev-4b | 61,8 |
| 106 | Kahn1 4B (Okura66, Qwen3.5-4B LoRA merge, author's sysone engine) | 61,7 |
| 107 | pplx-decider-flash-v1.2 (Perplexity, Qwen3.5-4B full fine-tune, marker readout, self-hosted) | 61,7 |
| 108 | lev (Interfaze AI, Qwen3.5-4B + LoRA r32, label-token readout + candidate-path head, per-bucket temperatures) | 61,7 |
| 109 | Decision 2.0 Lux 9B (vLLM-SR, Qwen3.5-9B lineage) | 61,6 |
| 110 | JEB-35B-A3B (szybkie.ai, Qwen3.6-35B-A3B LoRA) | 61,6 |
| 111 | Decisio v0.9.0 on Qwen3.6-35B-A3B (official FP8, self-hosted) | 61,5 |
| 112 | blink-4b (thegovind, Qwen3.5-4B derivative) | 61,4 |
| 113 | Hopper | 61,2 |
| 114 | Malkuth-4B (newfull5, Kev post-train) | 61,1 |
| 115 | Qwen3.5-9B Jev-like data-mix v2 | 60,9 |
| 116 | Decision 4B v1.2 (FlyMyJev, Qwen3.5-4B + LoRA) | 60,4 |
| 117 | jiwo 4b (Qwen3.5-4B backbone with a trained decision head, one forward pass, native probabilities, self-hosted) | 60,2 |
| 118 | Verdin-4B (Victor Bennett, Qwen3.5-4B + merged LoRA r16 + pointer head) | 59,6 |
| 119 | Wald 4B | 59,6 |
| 120 | Quyet-1.0-Medium (Chinh Nguyen, Qwen3.5-4B decoder, option-letter logits) | 59,2 |
| 121 | Decision-4B (Eval Engine / Chromia) | 59,1 |
| 122 | Clef-omni | 58,3 |
| 123 | Manchego v2.1 | 58,3 |
| 124 | Messier One v0.2 (Qwen3.5-4B fine-tune, one prefill, self-hosted) | 58,1 |
| 125 | Diffusion Jev (DiffusionGemma 26B-A4B on patched SGLang, 48-step diffusion readout, self-hosted) | 58,1 |
| 126 | jev-local (Qwen3.5-9B) | 57,5 |
| 127 | Decision 4B v1.1 (FlyMyJev, Qwen3.5-4B + LoRA) | 57,2 |
| 128 | JEV Qwen3.5-9B Base NVFP4 | 57,1 |
| 129 | iota-decide-4b (Macrocosmos, Clef joint-head 4B, self-hosted) | 57,1 |
| 130 | Uniphore-4B-Decider | 57,1 |
| 131 | Jebadiah 9B v2 (Frontier Infra, Qwen3.5-9B LoRA) | 56,9 |
| 132 | CoCo-Decision-4B (corners-ai, LoRA on Qwen3.5-4B, option-label logits of one pass, served by oh-my-jev) | 56,9 |
| 133 | APUS-OpenJev-v1-9B (merged bf16 checkpoint-3000 on Qwen3.5-9B, their own native runtime, full depth, self-hosted) | 56,9 |
| 134 | Decider-4B (roudone, NeoHorse 4B pointer model, author server, self-hosted) | 56,8 |
| 135 | typecastlm (Mikhail Gribov, Qwen3.5-4B computed head) | 56,5 |
| 136 | Bespoke Nimble 9B (Bespoke Labs) | 56,5 |
| 137 | pjev-2b (Kailune AI, frozen MiniCPM5-2B + merged LoRA r64, letter-logit readout, T=1 stub) | 56,5 |
| 138 | TypeCastLM 1.4.0 (Mikhail Gribov, Qwen3.5-4B computed head) | 56,4 |
| 139 | RYOTIDE-Qwen (csabag, Qwen3.5-4B stock weights) | 56,3 |
| 140 | Compass 0.2.0 (adimyth, Qwen3.5-4B LoRA) | 56,1 |
| 141 | Standard One 3B SH v4 (Standard Thinking) | 56,0 |
| 142 | Standard One 3B SH (Standard Thinking) | 55,6 |
| 143 | SemIf, formerly OpenJev (Qwen3.5-4B, TheoLeeCJ) | 55,4 |
| 144 | ArseneLupin v1.1 (IBOU-SEARCH, Qwen3.5-4B LoRA) | 55,1 |
| 145 | local-jev Qwen3.5-4B | 55,0 |
| 146 | Decision 2B (FlyMy.AI, v59) | 54,8 |
| 147 | Jez-large-1.0.0 (pragmaticfrontier, Gemma-4-E4B label-logit decision model, author server, self-hosted) | 54,4 |
| 148 | Noma (Blackdrome AI Labs, Qwen3.5-4B-Base derivative) | 54,3 |
| 149 | spark-s1-4b-v6 (Open Spark Jev, abhishek085) | 54,0 |
| 150 | Imajev-2B (mohit67890, Qwen3.5-2B LoRA) | 54,0 |
| 151 | RYOTIDE-Qwen9 | 52,5 |
| 152 | Arbiter-4B (Goddragon, NeoHorse 4B pointer model, author server, self-hosted) | 52,3 |
| 153 | DecisionNode Edge 0.9.2 | 52,1 |
| 154 | D1A on Gemma 4 E4B (JohnP1 d1a-e4b) | 51,1 |
| 155 | ClassOne Qwen 3.5 9B (Qwen3.5-9B backbone with ClassOne decision heads, schema-constrained single forward pass, self-hosted) | 50,5 |
| 156 | kev 8B (research preview) | 50,0 |
| 157 | Liquid AI d1-3B | 49,5 |
| 158 | ZeroEntropy zerank-2 | 49,4 |
| 159 | JADE (Qwen3.8-27B LoRA) | 49,2 |
| 160 | open-alternative-jev (Qwen3.5-4B, IkerMoel) | 49,1 |
| 161 | OpenSourceJev (Qwen3.5-4B Q4_K_M, native llama.cpp) | 48,6 |
| 162 | Open-Jev 2B (Zefan Cai) | 47,7 |
| 163 | EXAONE-4.0-1.2B-JEV v0.3 (unofficial fine-tune of LGAI-EXAONE/EXAONE-4.0-1.2B, native option-label softmax, two reads averaged, self-hosted) | 47,7 |
| 164 | Malkuth-2B (newfull5, Kev post-train) | 47,6 |
| 165 | DecisionNode Edge Mini 0.9.2 | 47,6 |
| 166 | Nemotron Diffusion 8B (pst2154, optimized vLLM) | 47,5 |
| 167 | jiwo 0.8b (Qwen3.5-0.8B backbone with a trained decision head, one forward pass, native probabilities, self-hosted) | 47,2 |
| 168 | decisor-4b | 45,7 |
| 169 | D1A on Gemma 4 E2B (JohnP1 d1a-e2b) | 45,5 |
| 170 | Qwen3-Reranker-4B | 44,7 |
| 171 | decider-2b (Mapika) | 44,6 |
| 172 | BAAI bge-reranker-v2-m3 | 44,1 |
| 173 | kev 4B (research preview) | 44,0 |
| 174 | Mixedbread mxbai-rerank-base-v2 | 43,9 |
| 175 | lev-350m (Franck Verrot, LFM2.5-350M) | 43,3 |
| 176 | Jez-small-1.0.0 (pragmaticfrontier, Qwen3.5-0.8B label-logit decision model, author server, self-hosted) | 43,2 |
| 177 | Certo v1 (AltSlate Labs) | 43,2 |
| 178 | PurpleMIST-Mini-1.0 (Olaverse) | 43,0 |
| 179 | Chon-1.0-12B | 43,0 |
| 180 | smalljev semantic-v9 | 43,0 |
| 181 | Vega-U (Qwen3.5-4B reader) | 42,6 |
| 182 | jul fast (usejul/jul-decision-minicpm5-2b: MiniCPM5-2B fine-tune with a pointer head, option probabilities from one pass, self-hosted via `jul serve`) | 42,6 |
| 183 | Krino-2B | 42,4 |
| 184 | Alibaba GTE Reranker ModernBERT-base | 41,7 |
| 185 | Jeetu Decider 2B | 41,7 |
| 186 | watt-flash-0.1 (Zaitgeist Labs, 140M mmBERT-small encoder with an option-marker scorer, one forward pass per request, self-hosted) | 41,6 |
| 187 | Decision Fast (FlyMy.AI, v53a) | 40,6 |
| 188 | openJev Verdict 1.4 | 40,4 |
| 189 | Quyet-1.0-Small-EN (Chinh Nguyen, ModernBERT-base encoder with fixed heads) | 39,5 |
| 190 | ClassOne Gemma 4 E4B (Gemma-4-E4B backbone with ClassOne decision heads, schema-constrained single forward pass, self-hosted) | 38,4 |
| 191 | ClassOne Qwen 3.5 4B (Qwen3.5-4B backbone with ClassOne decision heads, schema-constrained single forward pass, self-hosted) | 38,1 |
| 192 | OpenDecision (ModernBERT-large zero-shot) | 38,0 |
| 193 | verdict-small (Manavarya09, multilingual-e5-small 118M) | 37,6 |
| 194 | kev 0.6B (research preview) | 36,9 |
| 195 | Vega 4B | 36,6 |
| 196 | WaterSheep (Samrat Dutta, ModernBERT-base encoder with calibrated heads, served by the author's own `watersheep --serve`) | 35,5 |
| 197 | Raw Phi-4 mini direct logits | 35,3 |
| 198 | Raw Qwen3 4B Instruct 2507 direct logits | 34,6 |
| 199 | Vega 0.8B | 33,0 |
| 200 | kev 0.5B | 32,9 |
| 201 | Liquid AI d1-omni-600M | 32,8 |
| 202 | Quyet-1.0-Small (Chinh Nguyen, SEA-LION-ModernBERT-300M encoder with fixed heads) | 32,5 |
| 203 | Quyet-1.0-Tiny (Chinh Nguyen, mmBERT-small encoder (16 layers kept) with fixed heads) | 31,1 |
| 204 | Open Jev JSON Canvas (JoshuaSP) | 30,6 |
| 205 | Phocinae-Largha-150M-v1 | 30,2 |
| 206 | Raw Qwen3 8B direct logits | 30,1 |
| 207 | system-one (Qwen3-8B, Sean Goedecke) | 29,7 |
| 208 | openJev Verdict (heman10x, ModernBERT-base 151M) | 26,6 |
| 209 | Tacet Sonata (CodePawl, 144M packed encoder on mmBERT-small, option-marker softmax in one pass, in-process via the author's tacet package) | 24,2 |
| 210 | CLM-8B (Contrastive-LM, clm-latest) | 20,1 |
| 211 | ClassOne Qwen 3.5 2B (Qwen3.5-2B backbone with ClassOne decision heads, schema-constrained single forward pass, self-hosted) | 19,2 |
| 212 | Deem 0.8B v1 | 18,1 |
| 213 | Laya multilingual | 17,0 |
| 214 | ClassOne Gemma 4 E2B (Gemma-4-E2B backbone with ClassOne decision heads, schema-constrained single forward pass, self-hosted) | 15,2 |
| 215 | Julia-1 (Supersonic Labs) | 9,9 |
| 216 | Raw Qwen3 1.7B direct logits | 8,3 |
| 217 | Raw Qwen3 0.6B direct logits | 8,2 |
| hors classement | AJev lm6q (Gemma 4 31B) | 81,4 |
| hors classement | Aleria Judge 1.1.1 (ALERIA-ML, inference technique on stock Gemma 4 31B, author server, self-hosted) | 76,9 |
| hors classement | OpenJev (thinking, BF16) | 75,8 |
| hors classement | SimpleJev Qwen3.8-27B (self-hosted, v1.6 pool) | 74,9 |
| hors classement | Nemotron Deciffusion 8B v2 (pst2154) | 69,3 |
| hors classement | Open-Jev 27B v1.1 (Zefan Cai, LoRA + decision head on Qwen3.8-27B) | 68,0 |
| hors classement | LitJev (Qwen3.8-27B) | 66,6 |
| hors classement | Bev / Bonsai 27B | 66,1 |
| hors classement | Nagi-HUGE 12B (nagisanzenin, Gemma-4-12B LoRA) | 64,0 |
| hors classement | Open-Jev 9B (Zefan Cai) | 61,8 |
| hors classement | Kev-9B v2 (Jared Palmer, Qwen3.5-9B-Base LoRA + pointer head) | 61,5 |
| hors classement | APUS-OpenJev-v1-35B-A3B (merged bf16 checkpoint-5949 MoE on Qwen3.5-35B-A3B, their own native runtime, full depth, self-hosted) | 60,4 |
| hors classement | PurpleMIST-Flash-1.0 (Olaverse) | 60,1 |
| hors classement | Standard One 8B SH (Standard Thinking) | 60,0 |
| hors classement | reflex 4B (kshetrajna12) | 58,9 |
| hors classement | Standard One 8B (Standard Thinking) | 58,4 |
| hors classement | Bongard-mini (AgentBull, T5Gemma-2 4B-4B) | 54,8 |
| hors classement | RYOTIDE-Gemma (csabag, Gemma-4-E4B stock weights) | 54,5 |
| hors classement | Nev-2B (EryriLabs) | 49,9 |
| hors classement | Laya typed-decisions | 44,2 |
| hors classement | Qwen3.5-0.8B Decision Model (Mourad Ghafiri) | 42,6 |
| hors classement | Gutsy 0.8B v0.3 | 42,2 |
| hors classement | Reflex Instinct 0.6B (Atlas AI) | 38,6 |
| hors classement | jevora-9b-v1 (kadirbekovvv / doxanocap, Qwen3.5-9B-Base derivative) | 36,9 |
| hors classement | Vega-U (Qwen3.5-0.8B reader) | 36,1 |
| hors classement | open-jev-deberta-v3-large (local CPU) | 35,4 |
| hors classement | modernbert-jev v2 | 34,6 |
| hors classement | Laya (Convai Innovations, ModernBERT-large 421M) | 32,5 |
| hors classement | SimpleJev (Qwen3.5-0.8B, CPU) | 31,1 |
| hors classement | TDE general large v0.1 (tdelab, ModernBERT-large encoder) | 30,9 |
| hors classement | Mirror | 25,8 |
| sans rang | Torchcast Decision 31B v1.0 | 91,8 |
| sans rang | MetaDecision-31B-Rain | 90,6 |
| sans rang | WeMM-JEV (Rover00) | 88,7 |
| sans rang | Jeff-1.1-Large (jgeuter) | 88,2 |
| sans rang | VisionJEV 1.0.0 | 86,5 |
| sans rang | Piccolo 27B (agentscope-ai) | 86,0 |
| sans rang | oenogarum (plantwaterer; commercially restricted) | 83,9 |
| sans rang | allec-oavg (plantwaterer; commercially restricted) | 83,2 |
| sans rang | Decision-King v1.1 (xsdfasfgsa) | 82,2 |
| sans rang | Arcbright Elf 1 12B (r2) | 79,9 |
| sans rang | Jeff-1.1-Qwen-4B (jgeuter) | 79,6 |
| sans rang | Decisio 12B tuned commit-think (variant of the Decisio family) | 78,7 |
| sans rang | metask-jev-rain-12B | 78,3 |
| sans rang | H2O-Lightning-4B v1.4.1 | 77,0 |
| sans rang | Piccolo 9B (agentscope-ai) | 76,6 |
| sans rang | cribl-decision-1.0-27B (Cribl) | 76,5 |
| sans rang | Umpire 35B-A3B | 74,5 |
| sans rang | z3-mini-soup-preview (Zils AI) | 74,5 |
| sans rang | iota-decide-4b-v0.4 (Macrocosmos, letter-readout 4B, self-hosted) | 74,0 |
| sans rang | Piccolo 4B (agentscope-ai) | 73,6 |
| sans rang | Saif 4B (Gyr Technology) | 66,4 |
| sans rang | cribl-decision-1.0-9B (Cribl) | 65,9 |
| sans rang | cribl-decision-2 (Cribl) | 65,1 |
| sans rang | Granite Switch 4.1 30B (revision 4.1.1, decision adapter) | 64,5 |
| sans rang | Jev-Style-2B | 61,2 |
| sans rang | baselabs joey-1 (2B) | 57,8 |
| sans rang | Granite Switch 4.1 8B (revision 4.1.1, decision adapter) | 57,1 |
| sans rang | Sors-2B (SakuraYuyuko) | 56,3 |
| sans rang | Amber 4B (MLukauskas) | 56,1 |
| sans rang | cribl-decision-2-mini (Cribl) | 53,2 |

- Source : JevBench, Benchmark Heaven : https://benchmarkheaven.com/jev-models (relevé du 9 novembre 2026)

Source : QUELLE IA, édition du 9 novembre 2026, quelleia.com, d'après JevBench, Benchmark Heaven.