JevK5
Quadrant scores
See the full quadrantScored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.
- Maturity3.5/10
Apache-2.0 weights available; runtime published; JevBench v1.4.2 scores independently verified; no hosted inference or SLA.
- Capability6.1/10
Official boards (22 Oct): Decision Index 0.4.1 Full 36.59 (#130) for JevK5 v0.3; JevBench Intelligence 38.5, calibration 89.9, p50 0.019 s. Choice+Score via the jevk5 runtime.
- Adoption12/100
Brand-new release (23 Sep 2026). HF page up; community uptake unknown.
Vendor claims
- H100 latency ~9ms[Vendor claim — not independently verified]
Board audit of 17 Oct 2026: see the “Board audit” section below; where it conflicts with older board statements on this page, it wins.
JevK5 is an open-weight Jev alternative based on Qwen3.5-4B with LoRA merged. It uses a one-pass softmax readout — it does not generate text, so there is nothing to parse and nothing to hallucinate. Released under Apache-2.0.
JevBench v1.4.2 confirms #3 overall (62.04 score) and #1 among pure Apache-2.0 open-weight models — Decider-4B is also open but uses mixed licensing for some components.
A smaller JevK5-2B variant (Qwen3.5-2B) ships in the same family for lighter deployments (~3.5GB bf16).
Specs
| Attribute | Value |
|---|---|
| Author | alibiserikbay (independent) |
| Base model | Qwen3.5-4B + LoRA merged |
| License | Apache-2.0 |
| Launch | |
| Status | Open weights (Hugging Face) |
| Latency | ~9ms on H100 (UnverifiedClaim) |
| Pricing | Self-host only (free weights) |
| Decision types | Choice, Score (via jevk5 runtime) |
| Context | 32K (inherited from Qwen3.5) |
| Smaller variant | JevK5-2B — Qwen3.5-2B base |
| GGUF | JevK5-GGUF |
Architecture
JevK5 follows the one-pass softmax pattern: given state and typed questions, it produces probabilities over options without generating tokens. The jevk5 Python runtime handles the inference loop and exposes Choice, Score, and Noul question types (verify Noul support in the runtime docs).
JevBench v1.4.2 (verified)
| Metric | JevK5 | JevK5-2B | Rank |
|---|---|---|---|
| Overall | 62.04 | 56.48 | #3 / #9 |
| Intelligence | 48.6 | 46.2 | - |
| Calibration | 77.0 | 71.3 | - |
| Speed | 86.1 | 89.7 | - |
| Cost | 59.7 | 68.2 | - |
JevK5 holds #3 overall and #1 among pure Apache-2.0 open-weight models (Decider-4B is also open but uses mixed licensing for some components).
Fit / anti-fit
Fit when you need: self-hosted typed decisions, Apache-2.0 deployment, latency-sensitive routing on capable hardware.
Anti-fit when you need: hosted inference, production SLA, generation, or workflows that require free-text output.
JevBench v1.6.1 (Benchmark Heaven, checked 7 Oct 2026)
On the v1.6.1 open-weights board (full re-measure, not comparable with v1.5.7), JevK5 v0.3 (4B) is #20 at composite 37.4, capability 64.2 (#14). Intelligence 38.5 is below the board’s soft gate at 50, which pulls the composite down; Calibration is 89.9. JevK5 v0.2.0: 32.1 (#25), capability 62.8 (#17). See news. No catalog score change.
JevBench v1.5.7 (earlier edition, checked 5 Oct 2026; not comparable with v1.6)
JevK5 v0.3 (4B) is #4 on the live board at official score 71.9 (capability 72.3). That is a different edition from the v1.4.2 figure (62.04, #3) already on this page — do not mix them. Source: Benchmark Heaven; news. No catalog score change yet.
Board audit (17 Oct 2026)
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) | JevK5 9B v0.3.3 | Full 48.35 (#65, tie group 16); public 47.62 (#76); same-skills 48.19; new-domains 43.33 |
| Decision Index 0.3.2 (16 Oct 19:00 UTC) | JevK5 v0.3 | Full 41.54 (#91, tie group 22); public 41.14 (#95); same-skills 41.75; new-domains 37.73 |
| Decision Index 0.3.2 (16 Oct 19:00 UTC) | JevK5 Lite | Full 6.29 (#161, tie group 48); public 10.28 (#161); same-skills 10.37; new-domains 5.02 |
| JevBench open weights (live 17 Oct) | JevK5 v0.3 | Capability 64.2; Intelligence 38.5; calibration 89.9; p50 0.019 s; $0.017/1k |
| JevBench open weights (live 17 Oct) | JevK5 v0.2.0 | Capability 62.8; Intelligence 36.3; calibration 89.4; p50 0.040 s; $0.017/1k |
Board-status statements on this page to recheck against the table (may be superseded):
- “JevBench v1.4.2 confirms #3 overall (62.04 score) and #1 among pure Apache-2.0 open-weight models — Decider-4B is also open but uses mixed licensing f…”
Score update (audit, 17 Oct 2026). Capability 7.4 → 6.3, first full working (the earlier score had no stored breakdown) = types 6 × 0.30 + calibration 6 × 0.25 + latency 7 × 0.25 + accuracy 6 × 0.20 = 6.25 → 6.3. Types from decisionTypes; calibration from the JevBench axis 89.9 (≥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 0.019 s without p99 (≤0.25 s → 7, ≤0.5 s → 6, ≤1 s → 5, above → 3); accuracy Decision Index Full 41.54 ÷ 60.11 × 8 = 5.5, half-up. Maturity and adoption unchanged.
Hugging Face on 17 Oct (alibiserikbay/JevK5): 12,486 downloads, 23 likes, licence tag apache-2.0, last modified 2026-09-25.
Board audit (20 Oct 2026, DI 0.4)
Decision Index 0.4 (20 Oct; Full now from the private parts only) lists JevK5 v0.3 at Full 37.34 (#114); it was 41.54 on the archived 0.3.2 edition. Accuracy rule, reference Jev 1.13 on DI 0.4 Full (61.26): 8 × 37.34 ÷ 61.26 = 4.88 → 5 (was 6). Capability = 6 × 0.30 + 6 × 0.25 + 7 × 0.25 + 5 × 0.20 = 6.05 → 6.1 (was 6.3). Other sub-scores, maturity and adoption unchanged.
Board audit (22 Oct 2026, DI 0.4.1)
Decision Index 0.4.1 (22 Oct; reference Jev 1.13 Full 59.22, was 61.26 on 0.4) lists this page’s row at Full 36.59 (#130). Accuracy rule: 8 × 36.59 ÷ 59.22 = 4.94 → 5 (was 5). Capability unchanged. Maturity and adoption unchanged.
