ENPT

Jeffan-31b-1710

Last updated:

Open121–121ms$0/M input

Quadrant scores

See the full quadrant

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

  • Maturity4.3/10

    Apache-2.0 bf16 weights on google/gemma-4-31B-it, loadable like the base model, with a vLLM 0.29.0 recipe (B200) in the card. The card calls it a research model and first release. Independent author, no hosted API, no integrations, no SLA.

  • Capability5.3/10

    Decision Index 0.4.1 (22 Oct) Full 69.75 (#2, no tie), above Jev 1.13 (59.22). Multiple choice only, one question per prompt, up to 26 options (A–Z); no yes/no or score. No ECE in the card (the board measures 0.047); board median 121 ms.

  • Adoption1/100

    About 71 downloads and 6 likes three days after release (22 Oct). No integrations or production users found.

Vendor claims

publicado (OK do Fabricio 22/10 BRT). Scores are calculated from scratch.

What it is

Jeffan-31b-1710 (Hugging Face, Apache-2.0, released 19 Oct 2026) is a research model built on Gemma-4-31B-it. You give it a state (a document, a conversation, a list of tools) and a multiple-choice question; it returns a probability for every option from its next-token distribution over the option codes. No text is generated. Inputs up to 8,192 tokens.

Limits (read before you ship)

  • Multiple choice only, one question at a time. Each question needs its own prompt in the card’s fixed layout (State: / Question: / Options: with codes A, B, …). There is no yes/no (noul) or score type and no shared state across questions, so it is not a drop-in for /v1/systemone.
  • At most 26 options (A–Z). Other prompt formats “may differ”.

Board result

Board Row Result
Decision Index 0.4.1 (22 Oct 2026) Jeffan-31b-1710 Full 69.75 (#2, no tie); public 73.13 (#2); Jev 1.13 59.22 (#48), Jev 1.14 60.24

The card says “#1” (21 Oct, 69.8); in the 22 Oct data Fastino’s GLiDE V1.2 is first (71.52). Not checked against JevBench or Artificial Analysis; the boards are never combined.

Author claims (UnverifiedClaim)

  • “#1 on the Jev Decision Index leaderboard (21 Oct 2026)”, Full 69.8 of 220 models (source)

Scores (from scratch, 22 Oct 2026)

  • Capability 5.3 = types 3 × 0.30 + calibration 4 × 0.25 + latency 7 × 0.25 + accuracy 8 × 0.20 = 0.90 + 1.00 + 1.75 + 1.60 = 5.25 → 5.3. Types 3: choice only. Calibration 4: no ECE in the card. Latency 7: board median 121 ms. Accuracy: 8 × 69.75 ÷ 59.22 = 9.42, capped at 8.
  • Maturity 4.3 = availability 7 × 0.30 + docs 5 × 0.25 + integrations 2 × 0.25 + support 2 × 0.20 = 4.25 → 4.3.
  • Adoption 1 = engagement 1 × 0.35 + ecosystem 1 × 0.35 + production 0 × 0.30 = 0.70 → 1.