openJev-verdict-2.0

Quadrant scores

See the full quadrant

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

• Maturity3.5/10

  Day-one Apache weights + dashboard/WebGPU story, but brand-new independent release with no ops track record — maturity stays low.

• Capability5.5/10

  Official JevBench (17 Oct): Intelligence 4.1, calibration 49.1, p50 0.269 s. Full primitive set on ~150M ModernBERT with aggressive author accuracy/ECE claims; every Jev/Laya comparison remains UnverifiedClaim.

• Adoption12/100

  Loud README + dashboard in the S1Bench wave; still early community proof beyond the author channels.

Vendor claims

• UnverifiedClaim: top-1 ~77.10% on LocalLLaMA/typed-decisions held-out N=2000 (author vs Laya/Jev table)[Vendor claim — not independently verified]
• UnverifiedClaim: correctness-head ECE ~1.44% / ~20-25ms per decision (author)[Vendor claim — not independently verified]
• ModernBERT-base ~149.6M; Apache-2.0; WebGPU / GitHub Pages playground (author)[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.

openJev-verdict-2.0 is an independent ~150M ModernBERT-base decision model for typed Choice / Score / Noul (→ catalog boolean) in non-autoregressive passes — not TypeSafe.

Weights: heman10x/openJev-verdict-2.0; repo: github.com/Heman10x-NGU/openJev-verdict-2.0; dashboard: heman10x-ngu.github.io/openJev-verdict-2.0.

Limits (read before you ship)

• Every headline comparison vs Jev / Laya (accuracy, Brier, ECE, flip rate, “SOTA”) is author-reported → heavy UnverifiedClaim. ModelSystem.One has not audited the suite, splits, or Jev baseline protocol.
• Soft-label / dual-head story may not match your production correctness definition — fit gates on your labels.
• Marketing tone in the README is strong; use the card for orientation, not as a procurement scorecard.

Why it is in the catalog

Open encoder-scale System One–shaped I/O with a public playground, sitting in the same open cluster as Laya — with stricter claim hygiene than the README.

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.

Board	Row	Result
Decision Index 0.3.2 (16 Oct 19:00 UTC)	Verdict	Full 0 (#171, tie group 52); public 1.86 (#171); same-skills 2.41; new-domains 1.21
JevBench open weights (live 17 Oct)	openJev Verdict 1.4	Capability 40.4; Intelligence 5.0; calibration 75.7; p50 0.293 s; $0.0028/1k
JevBench open weights (live 17 Oct)	openJev Verdict	Capability 26.6; Intelligence 4.1; calibration 49.1; p50 0.269 s; $0.0028/1k

Score update (audit, 17 Oct 2026). Capability 6.9 → 5.5, first full working (the earlier score had no stored breakdown) = types 10 × 0.30 + calibration 3 × 0.25 + latency 6 × 0.25 + accuracy 1 × 0.20 = 5.45 → 5.5. Types from decisionTypes; calibration from the JevBench axis 49.1 (≥90 → 7, 85–89.9 → 6, 75–84.9 → 5, 60–74.9 → 4, below 60 → 3; no author ECE counted); latency from the board’s self-hosted p50 0.269 s without p99 (≤0.25 s → 7, ≤0.5 s → 6, ≤1 s → 5, above → 3); accuracy JevBench Intelligence 4.1 ÷ 63.6 × 8 = 0.5, half-up. Maturity and adoption unchanged.