← 首頁 / jev
JevCheese · Decision models
決策模型專區
Jev-class：輸入狀態與有界評分標準，輸出 typed 答案與機率分佈。榜序跟官方 JevBench Score；JevCheese 是四軸中位數的 Cheese 視角。大多數為開源／可自架——右側連結直達 GitHub 或 Hugging Face。
91Ranked
85Open / self-host
84With source link
v1.4.2.2Source rev
95 systems
| # | System | Open | Source | JevBench | JevCheese | I | C | S | K | $/1k dec. | Pub−seal |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Imajev-4B | 開源 | GitHub ↗ | 67.4 | 70.0 | 52.2 | 80.4 | 90.6 | 59.7 | $0.022 ~ | +49.1 |
| 2 | Plumb-4B (crh225, JevK5 v0.2 + LoRA) | 開源 | GitHub ↗ | 65.8 | 65.6 | 53.0 | 75.5 | 93.5 | 55.8 | $0.030 ~ | +51.6 |
| 3 | decider-4b v2 (Mapika) | 開源 | GitHub ↗ | 64.1 | 68.0 | 49.4 | 75.0 | 92.9 | 60.9 | $0.020 ~ | +48.8 |
| 4 | Jev 1.13.0 (TypeSafe AI) | 封閉 | Docs ↗ | 63.3 | 64.7 | 53.1 | 76.3 | 83.3 | 52.0 | $0.040 | +49.9 |
| 5 | JevK5 v0.2.0 | 開源 | GitHub ↗ | 62.0 | 67.0 | 48.9 | 74.5 | 91.1 | 59.5 | $0.022 ~ | +52.2 |
| 6 | Cygnet (blockbrain, frozen Gemma-4-12B-it) | 開源 | GitHub ↗ | 61.8 | 63.9 | 49.5 | 74.9 | 90.7 | 52.8 | $0.037 ~ | +54.1 |
| 7 | Hopper | 開源 | Hugging Face ↗ | 59.4 | 68.9 | 48.0 | 79.1 | 86.8 | 58.7 | $0.024 ~ | +48.2 |
| 8 | Winnow-12B Q8 | 開源 | Hugging Face ↗ | 55.6 | 58.9 | 48.3 | 64.8 | 82.3 | 52.9 | $0.037 ~ | +52.6 |
| 9 | reflex 4B (kshetrajna12) | 開源 | GitHub ↗ | 54.0 | 63.8 | 47.5 | 70.4 | 68.0 | 59.7 | $0.022 ~ | +51 |
| 10 | djev (Maisa, diffusion-gemma) | 開源 | GitHub ↗ | 52.2 | 56.5 | 47.0 | 55.4 | 91.4 | 57.6 | $0.026 | +54.1 |
| 11 | Jev-Omni (akhilaaa3, Gemma-4-12B merged) | 開源 | Hugging Face ↗ | 51.3 | 58.5 | 46.8 | 64.1 | 81.5 | 53.0 | $0.037 ~ | +56.6 |
| 12 | metask-jev-4b | 開源 | GitHub ↗ | 47.8 | 60.7 | 44.7 | 66.9 | 89.1 | 54.5 | $0.033 ~ | +52.1 |
| 13 | SemIf, formerly OpenJev (Qwen3.5-4B, TheoLeeCJ) | 開源 | GitHub ↗ | 47.7 | 63.1 | 44.4 | 66.8 | 83.7 | 59.5 | $0.022 ~ | +54.7 |
| 14 | Jobe Qwen3.5-4B (frozen) | 開源 | GitHub ↗ | 46.9 | 62.8 | 44.1 | 66.1 | 85.6 | 59.5 | $0.022 ~ | +55.3 |
| 15 | local-jev Qwen3.5-4B | 開源 | GitHub ↗ | 46.8 | 64.5 | 44.4 | 73.3 | 75.0 | 55.8 | $0.030 ~ | +54.5 |
| 16 | system-one-open (Gemma 4 E2B LoRA on an L4) | 開源 | GitHub ↗ | 45.1 | 59.8 | 44.2 | 54.9 | 77.0 | 64.8 | $0.015 ~ | +45.6 |
| 17 | spark-s1-4b-v6 (Open Spark Jev, abhishek085) | 開源 | GitHub ↗ | 44.6 | 52.7 | 45.1 | 47.6 | 81.0 | 57.9 | $0.025 ~ | +52.6 |
| 18 | Malkuth-4B (newfull5, Kev post-train) | 開源 | GitHub ↗ | 44.5 | 61.5 | 43.1 | 61.5 | 88.3 | 61.5 | $0.019 ~ | +51.5 |
| 19 | jqv (Qwen3-32B zero-shot) | 開源 | GitHub ↗ | 44.4 | 59.5 | 46.4 | 71.6 | 74.6 | 47.5 | $0.056 ~ | +51.8 |
| 20 | Qwen3-Reranker-4B | 開源 | Hugging Face ↗ | 43.5 | 57.2 | 44.6 | 65.2 | 78.7 | 49.2 | $0.050 | +38.1 |
| 21 | decider-35b-a3b (Mapika) | 開源 | Hugging Face ↗ | 41.2 | 56.2 | 47.2 | 65.3 | 80.8 | 45.3 | $0.067 ~ | +51.6 |
| 22 | Raw Qwen3 4B Instruct 2507 direct logits | 開源 | Hugging Face ↗ | 41.0 | 53.0 | 46.4 | 29.1 | 87.6 | 59.7 | $0.022 ~ | +42.4 |
| 23 | OpenSourceJev (Qwen3.5-4B Q4_K_M, native llama.cpp) | 開源 | GitHub ↗ | 40.9 | 62.1 | 41.8 | 60.3 | 82.0 | 64.0 | $0.016 ~ | +52.1 |
| 24 | ZeroEntropy zerank-2 | 開源 | Hugging Face ↗ | 40.2 | 62.8 | 42.1 | 75.8 | 79.0 | 49.8 | $0.047 | +41.6 |
| 25 | decision-machine-1 (milliseconds.ai) | 封閉 | milliseconds.ai ↗ | 39.9 | 61.0 | 41.3 | 68.3 | 92.9 | 53.7 | $0.035 | +41.9 |
| 26 | Malkuth-2B (newfull5, Kev post-train) | 開源 | GitHub ↗ | 38.9 | 57.5 | 41.3 | 53.5 | 91.4 | 61.5 | $0.019 ~ | +45.3 |
| 27 | Raw Phi-4 mini direct logits | 開源 | Hugging Face ↗ | 38.0 | 54.2 | 41.8 | 58.8 | 88.8 | 49.6 | $0.048 ~ | +36.6 |
| 28 | JEV Qwen3.5-9B Base NVFP4 | 開源 | Hugging Face ↗ | 37.7 | 57.3 | 46.8 | 67.7 | 93.3 | 43.3 | $0.077 ~ | +45.8 |
| 29 | OpenJev (DiffusionGemma 26B-A4B NVFP4, razorback16) | 開源 | GitHub ↗ | 36.9 | 50.2 | 45.4 | 55.0 | 83.2 | 45.5 | $0.066 ~ | +53.2 |
| 30 | kev 4B (research preview) | 開源 | GitHub ↗ | 36.1 | 51.9 | 42.1 | 39.6 | 75.7 | 61.8 | $0.019 ~ | +43.8 |
| 31 | Decision 2B (FlyMy.AI, v59) | 開源 | Hugging Face ↗ | 35.8 | 68.3 | 38.8 | 74.1 | 84.3 | 62.5 | $0.018 ~ | +49.4 |
| 32 | Qwen3.5-9B Jev-like data-mix v2 | 開源 | Hugging Face ↗ | 35.2 | 54.4 | 47.4 | 61.3 | 82.0 | 42.4 | $0.083 ~ | +49.1 |
| 33 | GPT-6 Luna (low reasoning effort) | 封閉 | — | 35.1 | 82.8 | 95.8 | 92.0 | 73.7 | 36.9 | $0.127 | +6.3 |
| 34 | SimpleJev Qwen3.8-27B | 開源 | GitHub ↗ | 34.6 | 61.4 | 51.6 | 74.5 | 71.2 | 39.5 | $0.104 ~ | +50.9 |
| 35 | NInfer Qwen3.8-Flash-Next mixed | 開源 | GitHub ↗ | 34.0 | 64.0 | 49.5 | 78.6 | 88.2 | 38.9 | $0.109 ~ | +55.5 |
| 36 | swanOne (blockbrain, Qwen3.8-Flash-Next NVFP4) | 開源 | GitHub ↗ | 33.6 | 61.8 | 52.6 | 71.0 | 82.5 | 38.6 | $0.111 ~ | +50.1 |
| 37 | GPT-6 Luna (default medium reasoning effort) | 封閉 | — | 33.3 | 83.0 | 97.4 | 93.5 | 72.6 | 36.0 | $0.135 | +4.1 |
| 38 | open-alternative-jev (Qwen3.5-4B, IkerMoel) | 開源 | GitHub ↗ | 33.2 | 59.2 | 38.6 | 58.7 | 83.5 | 59.6 | $0.022 ~ | +49.7 |
| 39 | jev-local (Qwen3.5-9B) | 開源 | GitHub ↗ | 32.5 | 54.7 | 45.2 | 64.2 | 69.2 | 43.3 | $0.077 ~ | +45.3 |
| 40 | Decision Fast (FlyMy.AI, v53a) | 開源 | Hugging Face ↗ | 32.5 | 70.7 | 37.1 | 65.3 | 81.6 | 76.1 | $0.0063 ~ | +37.6 |
| 41 | decider-2b (Mapika) | 開源 | Hugging Face ↗ | 30.7 | 52.2 | 38.5 | 43.5 | 83.2 | 61.0 | $0.020 ~ | +46.3 |
| 42 | jeff (Logan Markewich, GLiFormer 400M) | 開源 | GitHub ↗ | 30.6 | 65.7 | 36.8 | 67.9 | 63.5 | 76.6 | $0.0060 ~ | +29.7 |
| 43 | Laya (Convai Innovations, ModernBERT-large 421M) | 開源 | Hugging Face ↗ | 30.3 | 67.4 | 36.1 | 63.7 | 71.1 | 86.2 | $0.0029 ~ | +27.6 |
| 44 | Autoloops – Gemma 4 31B IT | 封閉 | — | 30.0 | 69.5 | 59.8 | 79.2 | 84.0 | 36.0 | $0.136 | +47.1 |
| 45 | Standard One 8B (Standard Thinking) | 開源 | Hugging Face ↗ | 29.1 | 56.0 | 46.2 | 65.9 | 92.0 | 39.4 | $0.104 ~ | +50 |
| 46 | lev-350m (Franck Verrot, LFM2.5-350M) | 開源 | GitHub ↗ | 28.5 | 73.3 | 34.8 | 70.6 | 85.3 | 76.1 | $0.0063 ~ | +33.4 |
| 47 | openjev-sglang (Qwen3.6-35B-A3B on SGLang) | 開源 | GitHub ↗ | 27.6 | 59.3 | 49.4 | 69.2 | 77.1 | 36.5 | $0.131 ~ | +52.2 |
| 48 | Von (wfzyx, Option-Marker 395M) | 開源 | GitHub ↗ | 27.5 | 73.1 | 34.5 | 75.7 | 70.5 | 77.8 | $0.0055 ~ | +29.2 |
| 49 | NInfer Qwen3.8-27B NVFP4 (T=1.5) | 開源 | GitHub ↗ | 26.9 | 63.7 | 51.5 | 76.0 | 80.1 | 35.2 | $0.145 ~ | +50 |
| 50 | NInfer Qwen3.8-27B NVFP4 | 開源 | GitHub ↗ | 26.3 | 59.3 | 51.5 | 67.2 | 80.1 | 35.2 | $0.145 ~ | +50 |
| 51 | kev 8B (research preview) | 開源 | GitHub ↗ | 25.6 | 42.9 | 41.8 | 40.2 | 74.9 | 44.0 | $0.073 ~ | +49.7 |
| 52 | JevOne | 開源 | Hugging Face ↗ | 25.5 | 62.6 | 47.6 | 77.6 | 88.5 | 35.9 | $0.137 ~ | +55.8 |
| 53 | typecastlm (Mikhail Gribov, Qwen3.5-4B computed head) | 開源 | GitHub ↗ | 25.3 | 61.1 | 34.0 | 62.1 | 92.6 | 60.2 | $0.021 ~ | +48.1 |
| 54 | SimpleJev Qwen3.6-35B-A3B | 開源 | GitHub ↗ | 24.9 | 52.8 | 45.7 | 59.8 | 75.0 | 38.1 | $0.116 ~ | +53.1 |
| 55 | kev 0.6B (research preview) | 開源 | GitHub ↗ | 24.8 | 62.8 | 34.2 | 50.0 | 75.6 | 76.1 | $0.0063 ~ | +42.6 |
| 56 | Raw Qwen3 8B direct logits | 開源 | Hugging Face ↗ | 23.7 | 43.8 | 45.7 | 24.1 | 86.3 | 41.9 | $0.087 ~ | +42.1 |
| 57 | system-one (Qwen3-8B, Sean Goedecke) | 開源 | GitHub ↗ | 23.4 | 42.5 | 43.6 | 32.8 | 84.4 | 41.5 | $0.089 ~ | +47.5 |
| 58 | OpenDecision (ModernBERT-large zero-shot) | 開源 | GitHub ↗ | 21.6 | 66.2 | 31.8 | 57.1 | 79.9 | 75.3 | $0.0066 ~ | +27.6 |
| 59 | LitJev (Qwen3.8-27B) | 開源 | GitHub ↗ | 19.5 | 56.5 | 46.3 | 76.6 | 66.7 | 33.6 | $0.163 ~ | +55.3 |
| 60 | openJev Verdict 1.4 | 開源 | Hugging Face ↗ | 19.0 | 75.1 | 29.4 | 72.0 | 78.1 | 82.4 | $0.0039 ~ | +29.7 |
| 61 | kev 0.5B | 開源 | GitHub ↗ | 18.9 | 62.9 | 30.5 | 49.7 | 77.0 | 76.1 | $0.0063 ~ | +22.1 |
| 62 | Bespoke Nimble 9B (Bespoke Labs) | 開源 | GitHub ↗ | 18.7 | 51.4 | 46.3 | 56.4 | 78.7 | 33.4 | $0.166 ~ | +50.8 |
| 63 | GPT-5.6 Luna (low reasoning effort) | 封閉 | — | 18.5 | 82.5 | 93.1 | 87.4 | 77.5 | 28.5 | $0.242 | +8.4 |
| 64 | openJev Verdict (heman10x, ModernBERT-base 151M) | 開源 | GitHub ↗ | 18.1 | 61.8 | 30.0 | 47.0 | 76.7 | 83.1 | $0.0037 ~ | +30.7 |
| 65 | Raw Qwen3 1.7B direct logits | 開源 | Hugging Face ↗ | 18.1 | 49.1 | 33.2 | 24.3 | 89.7 | 64.9 | $0.015 ~ | +28.1 |
| 66 | reflex-27b (Qwen3.8-27B) | 開源 | GitHub ↗ | 17.8 | 56.9 | 46.4 | 77.2 | 67.5 | 32.3 | $0.181 ~ | +57.5 |
| 67 | JevAct (einptein, jev1-2b-v2) | 封閉 | GitHub ↗ | 16.9 | 59.7 | 29.3 | 55.1 | 76.5 | 64.3 | $0.015 ~ | +38.5 |
| 68 | djev (thinking) | 開源 | GitHub ↗ | 15.2 | 73.4 | 71.6 | 87.8 | 75.2 | 26.9 | $0.274 ~ | +27.4 |
| 69 | GLiNER2 large (Fastino) | 開源 | Hugging Face ↗ | 15.1 | 46.4 | 31.1 | 24.8 | 61.7 | 73.3 | $0.0077 ~ | +28.1 |
| 70 | OpenJev (thinking, BF16) | 開源 | GitHub ↗ | 14.8 | 58.1 | 58.1 | 58.1 | 76.1 | 27.8 | $0.255 ~ | +46.5 |
| 71 | Qwen3.5-0.8B Decision Model (Mourad Ghafiri) | 開源 | Hugging Face ↗ | 14.5 | 58.7 | 28.1 | 68.2 | 49.2 | 75.7 | $0.0065 ~ | +24.6 |
| 72 | Gemini 3.1 Flash-Lite | 封閉 | — | 14.3 | 56.9 | 54.5 | 59.3 | 81.8 | 27.4 | $0.264 | +48.4 |
| 73 | open-jev-deberta-v3-large (local CPU) | 開源 | GitHub ↗ | 12.7 | 66.3 | 25.6 | 66.6 | 66.0 | 74.0 | $0.0073 ~ | +22.8 |
| 74 | smalljev semantic-v9 | 開源 | GitHub ↗ | 12.3 | 58.5 | 25.7 | 59.2 | 79.8 | 57.9 | $0.025 ~ | +33.7 |
| 75 | GLiNER2 (Fastino, gliner2.5-base) | 開源 | GitHub ↗ | 11.8 | 49.6 | 27.4 | 25.2 | 71.8 | 83.1 | $0.0037 ~ | +28.8 |
| 76 | Instinct (ZooWork, Qwen3.8-27B) | 封閉 | GitHub ↗ | 11.4 | 64.9 | 51.1 | 78.7 | 83.9 | 24.6 | $0.325 ~ | +51.5 |
| 77 | Open-Jev 9B (Zefan Cai) | 開源 | GitHub ↗ | 11.2 | 53.0 | 44.2 | 61.8 | 72.0 | 28.1 | $0.249 ~ | +47.6 |
| 78 | Open-Jev 2B (Zefan Cai) | 開源 | GitHub ↗ | 10.0 | 48.8 | 42.3 | 55.3 | 73.5 | 28.1 | $0.249 ~ | +38.2 |
| 79 | GLiNER2.5 multi (Fastino, 287M) | 開源 | Hugging Face ↗ | 9.8 | 62.5 | 23.1 | 57.2 | 67.8 | 82.4 | $0.0039 ~ | +16.1 |
| 80 | CLM-8B (Contrastive-LM, clm-latest) | 開源 | GitHub ↗ | 8.6 | 59.1 | 22.4 | 39.8 | 93.6 | 78.4 | $0.0052 ~ | +16.7 |
| 81 | SimpleJev (Qwen3.5-0.8B, CPU) | 開源 | GitHub ↗ | 7.5 | 53.3 | 21.5 | 49.1 | 57.5 | 68.3 | $0.011 ~ | +19.8 |
| 82 | GLiNER2.5 small (Fastino, 74M) | 開源 | Hugging Face ↗ | 7.2 | 64.3 | 20.5 | 50.7 | 77.8 | 82.4 | $0.0039 ~ | +17.3 |
| 83 | Raw Qwen3 0.6B direct logits | 開源 | Hugging Face ↗ | 7.1 | 48.4 | 22.8 | 20.7 | 89.9 | 73.9 | $0.0074 ~ | +22.4 |
| 84 | verdict-small (Manavarya09, multilingual-e5-small 118M) | 開源 | GitHub ↗ | 5.7 | 76.0 | 18.1 | 66.9 | 85.0 | 96.6 | $0.0013 ~ | +24.8 |
| 85 | DeepSeek V4.1 Flash (thinking default) | 開源權重 | Hugging Face ↗ | 4.8 | 82.8 | 94.0 | 95.5 | 71.6 | 16.8 | $0.594 | +3 |
| 86 | Mirror | 開源 | GitHub ↗ | 2.1 | 48.4 | 13.6 | 26.0 | 70.8 | 73.3 | $0.0077 ~ | +32 |
| 87 | Mixedbread mxbai-rerank-base-v2 | 開源 | Hugging Face ↗ | 0.4 | 76.0 | 6.8 | 84.1 | 87.5 | 67.9 | $0.012 | +2.8 |
| 88 | BAAI bge-reranker-v2-m3 | 開源 | Hugging Face ↗ | 0.2 | 78.8 | 5.0 | 84.2 | 89.5 | 73.4 | $0.0077 | +11.5 |
| 89 | Alibaba GTE Reranker ModernBERT-base | 開源 | Hugging Face ↗ | 0.1 | 74.3 | 4.8 | 78.9 | 90.6 | 69.6 | $0.010 | +0.3 |
| 90 | Certo v1 (AltSlate Labs) | 開源 | Hugging Face ↗ | 0.0 | 88.5 | 0.1 | 83.0 | 94.0 | 100.0 | $0.0010 ~ | +2.1 |
| 91 | Open Jev JSON Canvas (JoshuaSP) | 開源 | GitHub ↗ | 0.0 | 46.9 | 48.2 | 0.0 | 84.1 | 45.6 | $0.065 ~ | +53.2 |
| — | classifier.dev (fast tier) | 封閉 | classifier.dev ↗ | 70.8 | 78.4 | 51.6 | 72.4 | 87.6 | 84.3 | $0.0033 ~ | +50.9 |
| — | Qwen3.8 27B (Chutes TEE) | 開源權重 | — | 0.0 | 50.8 | 40.4 | 93.6 | 61.3 | 0.0 | $2.669 ~ | +50.1 |
| — | Needle 3, options as tools (post-hoc adapter mode) | 開源 | GitHub ↗ | — | — | — | — | — | — | $0.014 ~ | — |
| — | Needle 3 (Cactus, 2-bit, local CPU) | 開源 | GitHub ↗ | — | — | — | — | — | — | $0.024 ~ | — |
I/C/S/K = Intelligence / Calibration / Speed / Cost. Cost is USD per 1,000 decisions, not per 1,000 tokens. Attribution: Benchmark Heaven JevBench (MIT harness); not merged into CF-1.3. · 榜面計分 2026-09-27T18:49:51+00:00 · 本站抓取 2026-09-28T08:01:09Z