Jev directory
CommunityJev Models & Reimplementations
Kenning-XL v0.6
An experimental decoder-based System One engine: Qwen3-1.7B with a constrained answer-token readout that answers the /v1/systemone wire format and can escalate to a short reasoning trace, Apache-2.0.
- Reading time
- 2 min
- Published by
- Community
- Author
- SystemOne Dev
- Last verified
- Oct 2026
Used inJev for Software & DevTools
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- Macro 0.688 across 1,328 held-out items, against the v0.5 cross-encoder's 0.653, Clef's 0.791, and Jev's 0.830.
- Strongest on agent decisions at 0.860, beating Clef's 0.793, but trails the cross-encoder on logs (0.440 vs 0.520).
- Deliberate mode more than doubles the daily-limit arithmetic task, 0.42 to 0.81, and transfers to held-out tasks.
- Served by SystemOne Builder's kenning-xl engine, which escalates to deliberate mode only for low-confidence, short, or structured questions.
- Merged LoRA on Qwen3-1.7B in bf16 with a 1,536-token context; not trained on any TypeSafe output.
Watch out
Apache-2.0 weights on an Apache-2.0 base, but the card labels it experimental: it does not match Clef overall (0.688 against 0.791 macro) and trails even the v0.5 cross-encoder on logs, quality, and text. It is not a plain transformers classifier and must be loaded through the builder, and its comparison numbers are the card's own.
Questions about this entry
Short answers drawn from the linked project or page.
It is not a plain transformers classifier: the card says to serve it through SystemOne Builder's kenning-xl engine (systemone_builder.kenning.xl_serve), which manages one-pass and deliberate modes.
It generates a short reasoning trace and then reads the answer, for numeric, policy, and table decisions a single pass cannot do; the serving engine escalates to it only for low-confidence, short, or structured questions.
Macro 0.688 against 0.653, led by agent decisions at 0.860 against 0.727; it trails on logs (0.440 against 0.520), text (0.752 against 0.756), and quality (0.439 against 0.479).
No. The card's comparison table puts Clef at 0.791 and Jev at 0.830 macro against Kenning-XL's 0.688; it does beat Clef on agent decisions, 0.860 against 0.793, and ties conversation at 1.000.
A merged LoRA fine-tune of Qwen3-1.7B in bf16 with a 1,536-token context, trained by readout distillation on the v0.5 structured corpus plus gold-trace deliberate training whose reasoning chains are rule-generated.
No. The card states it is not trained on any TypeSafe output; it implements a wire format compatible with TypeSafe AI's System One API and says it is not affiliated with or endorsed by TypeSafe AI.
The hardest arithmetic task, daily-limit checks, rose from 0.42 to 0.81 and the gains transfer to held-out tasks; logs, quality, and text are not trace-trained yet, which is why it trails there.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Standard One SH (StandardThinking)
LinkedIn: What multimodal decision models actually change, and what they do not.
LinkedInJevDirectory.org
Hacker News: Decision Index 0.2.1 scores open reproductions of Jev on a frozen suite.
jackbravo4 pointsDecision Index 0.2.1 – 70 open reproductions of TypeSafe Jev's Decision ModelRead the thread on Hacker NewsHacker News: Pointer to a DiffusionGemma plus vLLM tutorial for a Jev-shaped classifier.
1 pointHow to build a Jev-style classifier with DiffusionGemma and vLLMRead the thread on Hacker NewsMore Jev resources like this
A pointer-network decision model from Keyush Nisar that reads one agent state document plus typed questions and returns calibrated tool-routing probabilities in a single forward pass, reporting 89.00% zero-shot tool routing.
Jev Models & Reimplementations#community#open-models#local
Community
A LoRA plus 24-slot decision head over Qwen3.8-27B from Ming Xu's train-jev project: it serves TypeSafe's System One API with 0.851 Kev transfer-v4 accuracy and ECE 0.053, and needs an 80 GB GPU.
Jev Models & Reimplementations#community#open-models#local
Community
A 24 GB-GPU alternative from the train-jev project: Qwen3.5-4B plus a rank-16 LoRA and 24-slot decision head scoring 0.647 PK accuracy, 40.4 Decision Index and 43 ms p50 latency, ahead of Kev-4B at the same size.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune for coding decisions: roughly 9k rows from SWE-Gym teach keep/drop file relevance, priority scoring and injection detection, reporting 0.833 accuracy on 1,150 held-out rows at 126 ms p50.
Jev Models & Reimplementations#community#open-models#local
Community
A Laya fine-tune from Convai Innovations on the 1,200-case typed-decisions benchmark: it reports 0.777 accuracy on the official 400-case test set, above TypeSafe Jev 1.13.0 at 0.727 and the teacher ceiling at 0.735.
Jev Models & Reimplementations#community#open-models#local
Community
A 9B bilingual Turkish and English Jev-class decision model from Berkay Hasret: a 173 MB LoRA adapter on Qwen/Qwen3.5-9B that answers up to 16 typed questions about one text in a single forward pass and reports 0.783 macro accuracy on full JevBench.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.