Jeff 1

Quadrant scores

See the full quadrant

Scored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.

• Maturity3.2/10

  Apache LoRA + train-your-own docs on day one; brand-new Gestalt Lab release without hosted SLA — low maturity.

• Capability5.7/10

  Three primitives via first-token / sequence scoring on a 4B instruct; quality vs Jev is author-only (UnverifiedClaim).

• Adoption5/100

  Just published in the S1Bench gap set; public footprint still minimal beyond Gestalt Lab channels.

Vendor claims

• LoRA on Qwen/Qwen3-4B-Instruct-2507; Apache-2.0 adapter (author card)[Vendor claim — not independently verified]
• UnverifiedClaim: on 9730 FEVER-family labels, Jeff 1 acc 0.8183 vs live Jev 1.13.0 0.8283 (author)[Vendor claim — not independently verified]
• UnverifiedClaim: author ECE 0.0807 vs Jev 0.0932 on the same fact-check suite[Vendor claim — not independently verified]

publicado (OK do Fabricio 17/10 BRT). Board audit of 17 Oct 2026: see the “Board audit” section below; where it conflicts with older board statements on this page, it wins.

Jeff 1 is an open LoRA adapter on Qwen3-4B-Instruct-2507 for Jev-shaped typed decisions — not TypeSafe and not a trained-from-scratch System One base.

It scores caller-supplied labels with first-token or whole-sequence logprobs for choice, noul (→ catalog boolean), and score. Weights: GestaltLabs/Jeff-1; code + train guide: github.com/Gestalt-Lab/jeff.

Limits (read before you ship)

• Head-to-head numbers vs Jev 1.13.0 on FEVER / VitaminC / SciFact / Climate-FEVER are author-reported → UnverifiedClaim. ModelSystem.One has not reproduced them.
• Token-scoring LoRA ≠ pointer-readout / RLCD stack; API-ish shapes do not imply identical judgements.
• Confidence = max label probability — not a separately trained correctness head.
• Brand-new (2026-09-19); expect sharp edges in packaging and eval hygiene.

Why it is in the catalog

Public LoRA + typed decision I/O in the open-Jev wave. Catalogued beside Kev as another lora-adapter peer; place it well below hosted Jev until third-party benches exist.

Board audit (17 Oct 2026)

publicado (OK do Fabricio 17/10 BRT). Checked against the official boards on 17 Oct 2026: Decision Index 0.3.2 (data of 16 Oct, 19:00 UTC), JevBench and ImageJevBench live. The boards are separate sources and are never averaged. Older board statements and rankings further down this page are superseded where they conflict with this section.

Rows with similar names on the boards. The repo links differ from this page’s repo, so they are not attributed to this model:

Board	Row	Result
Decision Index 0.3.2 (16 Oct 19:00 UTC)	jeff GLiFormer-large	Full 7.71 (#160, tie group 47); public 8.27 (#163); same-skills 6.86; new-domains 13.92
JevBench open weights (live 17 Oct)	Jeff-1.0-Large	Capability 81.5; Intelligence 71.9; calibration 91.1; p50 0.080 s; $0.049/1k

Version / family note. The board rows named Jeff-1 / jeff GLiFormer-large point to different repositories from this page’s (Gestalt-Lab/jeff), so no board row is claimed for it. Score unchanged (5.7).

Hugging Face on 17 Oct (GestaltLabs/Jeff-1): 2,237 downloads, 11 likes, licence tag apache-2.0, last modified 2026-09-19.