Benchmarks
More
OverviewBenchmaxxingCompareChartsCost vs CapabilityProviders per ModelProvider explorerGatewaysEU & SovereignJev-class modelsImage JevBenchAbout
JevBench by Benchmark Heaven · v1.4.2.2 · individual system

Malkuth-2B (newfull5, Kev post-train)

Jev rebuild · by newfull5 (dhtocks)

JevBench v1.4.2.2 score

38.9

Rank #26 of 91 ranked systems.

24.4 points behind Jev 1.13.0's 63.3.

Where it sits among the 91 ranked systems. The marked tick is Jev 1.13.0 (63.3).

Published axes

intelligence
41.3
calibration
53.5
speed
91.4
cost
61.5

Against Jev 1.13.0

Four radars compare this fixed pair across the score axes, accuracy per tier, and accuracy by family on the hard tier and sealed set. Further out is better on every spoke.

  • A: Malkuth-2B — Jev rebuild · Score 38.9 (#26)
  • B: Jev 1.13.0 — Jev (TypeSafe, closed) · Score 63.3 (#4)

The four score axes

Radar: the four score axes, two systemsThe four score axes, Malkuth-2B vs Jev 1.13.0. Intelligence: 41.3 vs 53.1; Calibration: 53.5 vs 76.3; Speed: 91.4 vs 83.3; Cost: 61.5 vs 52.0.50100Intelligence41.3 · 53.1Calibration53.5 · 76.3Speed91.4 · 83.3Cost61.5 · 52.0
0–100, the values in the table. A label-only system has no calibration (counted as 0).

Accuracy per tier, incl. sealed

Radar: accuracy per tier, incl. sealed, two systemsAccuracy per tier, incl. sealed, Malkuth-2B vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 89% vs 99%; Judge: 86% vs 95%; Hard: 45% vs 74%; Sealed: 24% vs 37%.50100Easy100% · 100%Standard89% · 99%Judge86% · 95%Hard45% · 74%Sealed24% · 37%
Share correct per tier; Sealed = the 308 private decisions, aggregate only.

Current question set by family (hard + sealed)

Radar: current question set by family (hard + sealed), two systemsCurrent question set by family (hard + sealed), Malkuth-2B vs Jev 1.13.0. Ambiguous / abstain: 16% vs 43%; Judge: 45% vs 54%; Long policy: 27% vs 44%; Multi-hop: 30% vs 64%; Probability: 27% vs 63%; Temporal / numeric: 16% vs 28%; Trade-off: 29% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 75% vs 83%; Paraphrase: 43% vs 64%; Safety judge: 31% vs 38%.50100Ambiguous /abstain16% · 43%Judge45% · 54%Long policy27% · 44%Multi-hop30% · 64%Probability27% · 63%Temporal /numeric16% · 28%Trade-off29% · 55%Routing100% · 100%Trap /adversarial75% · 83%Paraphrase43% · 64%Safety judge31% · 38%
Share correct per family across the 220 hard-tier decisions (public and held out) and the 308 sealed decisions of v1.4, pooled; Routing is hard-tier only, Paraphrase and Safety judge sealed only.

Sealed set by family

Radar: sealed set by family, two systemsSealed set by family, Malkuth-2B vs Jev 1.13.0. Ambiguous / abstain: 11% vs 30%; Judge: 34% vs 34%; Long policy: 28% vs 28%; Multi-hop: 13% vs 45%; Paraphrase: 43% vs 64%; Probability: 25% vs 50%; Safety judge: 31% vs 38%; Temporal / numeric: 18% vs 29%; Trade-off: 27% vs 38%; Trap / adversarial: 50% vs 42%.50100Ambiguous /abstain11% · 30%Judge34% · 34%Long policy28% · 28%Multi-hop13% · 45%Paraphrase43% · 64%Probability25% · 50%Safety judge31% · 38%Temporal /numeric18% · 29%Trade-off27% · 38%Trap /adversarial50% · 42%
Share correct within each sealed family — system-level aggregates; the items stay private.

Availability and evidence

Openness
Code and weights marked open in the published row
License note
CC-BY-NC-4.0, research use only (XNLI and RACE in the training mix)
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our evaluator-owned Lium GPU pod (RTX PRO 6000), offline read-only container, author's server on loopback
Note on this row
Dhtocks/malkuth-2b rev 401304b989451876070d83c488271b4f927d03ab (rank-16 LoRA + pointer head over empero-ai/Qwen3.8-2B-Distill rev e37a2dc4acc68ad75a91e07e63168cb04cc06345, fitted T=1.1755), served by kev.serve from jaredpalmer/kev 557598fced1dada75dfbf36ed144dce309ac6ceb. Offline self-hosted inference of all 842 decisions (534 frozen v1.2 + 308 sealed v1.4) in a network-disabled, read-only container on an evaluator-owned Lium RTX PRO 6000 pod, through JevBench's unchanged typesafe adapter against the author's own server on loopback; no operator endpoint; no golds were exposed. Latency is the serial request wall time on the standard+judge items with the self-hosted adjustment. Cost is a labelled estimate: DeepInfra Qwen/Qwen3.5-4B list price $0.03/M input (no hosted ~2B listed; the 4B price errs high, the decider-2b precedent), read 2026-09-24, over the server's own usage.input_tokens; it is not a GPU bill.
Published source
https://github.com/newfull5/malkuth

From the public v1.4.2.2 aggregate. Scores and ranks can change when a new release is published.

Read the full board and published method. The overall score is a composite, not raw accuracy.