| # | Model | GitHub repo | GitHub stars | Paper | Backbone | Params | How SFT is done | RL | Curriculum | Mixes original data | Context window | Training data |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Kev | jaredpalmer/kev | 9,621 | Qwen3.5-0.8B/4B/9B-Base | 0.8B / 4B / 9B | LoRA | None | basics → dates/missing evidence → long docs → hard + dev tools |
Replay | 7,552 | 30.5k (5 subsets) | |
| 2 | NanoJev | TianyuCodings/NanoJev | 2,563 | System1-Arena (benchmark, same author) | Qwen3-0.6B | 0.6B | Full | None | None | Replay | 8,192 | 10,893 game decisions |
| 3 | Nimble original | bespokelabsai/nimble | 2,257 | Qwen3.5-9B | 9B | LoRA | None | None | None | 2,048 | 2,676 synthetic contrastive pairs | |
| 4 | jevlike | vinnylarouge/jevlike | 1,358 | TinyScorer (custom) | 40k | Full | None | imitation → DAgger |
Replay | 192 bytes | Synthetic menus, wiki hopping, chess, Doom | |
| 5 | rizzo-flow | Rizzo-AI-Academy/rizzo-flow | 951 | Spark-X2.5-4B | 4B | LoRA | None | None | General | 2,048 | 28,321 questions | |
| 6 | Strands v21 | strands-labs/strands-decider | 721 | Qwen3.5-2B-Base | 2B | LoRA | None | None | General + Replay + KL to base | 3,072 | Public sets + synthetic, Qwen3.5-4B teacher | |
| 7 | Open-Jev 2B / 9B | Zefan-Cai/Open-Jev | 435 | Qwen3.5-2B / 9B | 2B / 9B | LoRA | None | None | None | 4,096 | 80,816 synthetic rows | |
| 8 | imajev | mohit67890/imajev | 413 | Tech report (not arXiv) | Qwen3.5-2B/4B/9B | 2–9B | LoRA | None | human-labeled → teacher-labeled → hard → soft |
General + Replay | 4,096 | 504k human-labeled + teacher-labeled + synthetic hard |
| 9 | reflex | kshetrajna12/reflex | 174 | Qwen3.5-4B | 4B | LoRA | None | None | General | 4,096 | 6.4k–7.9k public mix | |
| 10 | Verdict-open-jev | Heman10x-NGU/Verdict-open-jev | 113 | ModernBERT-base + GLiClass, encoder | 151M | Full | None | None | None | 512 | 2,300 Banking77 | |
| 11 | Chinese Decision | yanqiangmiffy/chinese-decision | 53 | arXiv 2610.23170 | Qwen3.5-0.8B/2B/4B | 0.8–4B | LoRA | None | None | General | 16,384 | 13.05M decisions |
| 12 | OpenJev-RLCD | ZimmyGao/openjev-rlcd | 45 | arXiv 2609.38850 | Qwen3-1.7B | 1.7B | Full | Yes | None | KL to stage-1 model | n/s | MMLU-Pro 4k + GSM8K-Verify 4k |
| 13 | system-one-open | mithalouni/system-one-open | 41 | Gemma-4-E2B-it / Gemma-3-270M-it | 2B / 270M | LoRA (E2B) / Full (270M) | None | None | General | 2,048 | 92 public sets + synthetic | |
| 14 | Eikos-27B | caiovicentino/eikos | 41 | Qwen3.8-27B | 27B | LoRA | None | None | General | 14,000 | 22,216 rows, GLM-5.3-Flash teacher | |
| 15 | Bekko 17M/68M/400M | hotchpotch/bekko-system-one | 37 | Ettin-reranker, encoder | 17M / 68M / 395M | Full | None | None | None | 4,096 | 8.42M judgments (153 subsets) | |
| 16 | eve-rlcd | anthony-maio/eve-rlcd | 18 | Qwen3-0.6B-Base | 0.6B | Full | Yes | None | None | 512 | 64k public + synthetic | |
| 17 | jevbetter | olanotolu/jevbetter | 16 | Char n-gram + 2-layer Transformer (custom) | 8.9M | Full | None | None | None | 512 features | 4,000 synthetic | |
| 18 | Rev | robbalian/rev | 10 | Qwen3.5-4B/9B, Qwen3.8-27B | 4B / 9B / 27B | LoRA | None | None | General | 4,096 | 19,792 public decisions | |
| 19 | rlcd-lite | arnabgho/rlcd-lite | 10 | Qwen2.5-0.5B-Instruct | 0.5B | None | Yes | None | None | n/s | 800 synthetic tickets | |
| 20 | Jebadiah 27B / 9B | getainode/jebadiah | 10 | Qwen3.8-27B / Qwen3.5-9B | 27B / 9B | LoRA | None | None | General | 2,048 | 11,013 public items | |
| 21 | system-one-gemma | akash-kamat/system-one-gemma | 7 | Gemma-3-270M-Base | 268M | LoRA | None | None | General | 256 | 12,913 public questions | |
| 22 | sokudan | hiroki-abe-58/sokudan | 6 | ModernBERT-ja-310M, encoder | 315M | Full | None | None | None | n/s | 67,394 synthetic Japanese rows | |
| 23 | WebJev | lexmount/WebJev | 5 | Qwen3.5-35B-A3B-Base | 35B (3B active) | Full (routed experts frozen) | None | None | General | 16,384 | 3.19M mixed (incl. 1.24M knowledge MCQ) | |
| 24 | verdict-small | Manavarya09/verdict | 5 | e5-small, encoder | 118M | Full | None | None | None | 128 | 134,680 public rows | |
| 25 | Intelif | SkAndMl/intelif | 2 | Qwen3-4B | 4B | LoRA | None | None | General | 4,096 | 200,612 public rows + agent trajectories | |
| 26 | Certo | AltSlate-Labs/certo | 2 | Tech report (not arXiv) | ModernBERT-large, encoder | 395M | Full | None | None | None | 64 | 60k synthetic |
| 27 | Jevstral | AmirBraham/jevstral | 1 | Ministral-3-8B-Base | 8B | LoRA | None | basics → dates/missing evidence → long docs → hard + dev tools |
Replay | 8,192 | 30.5k (Kev data) | |
| 28 | nano-jev | shubham10divakar/nano-jev | 1 | MiniLM-L12, encoder | 33.4M | Full | None | None | None | 512 | 49,760 public QA/NLI | |
| 29 | System X v4 | Turinglogicuk/system-x | 1 | arXiv 2610.25456 | Qwen3-1.7B | 1.7B | LoRA | None | None | General + KL to base | n/s | 24,256 rows, Qwen3-14B teacher |
| 30 | jevkit | soumyacodes007/jev-compatible-model-training | 0 | Qwen3.5-4B | 4B | LoRA | None | None | None | 2,048 | Public classification/NLI | |
| 31 | saihai | hotchpotch/saihai | 0 | llm-jp-4.1-8B-thinking | 8B | LoRA | None | None | General | 8,192 | 33,532 rows, LLM-jp-32B teacher |