Benchmarks
More
OverviewBenchmaxxingCompareChartsCost vs CapabilityProviders per ModelProvider explorerGatewaysEU & SovereignJev-class modelsAbout
JevBench by Benchmark Heaven · v1.4.1 · individual system

Raw Qwen3 8B direct logits

This detail uses the public, hash-checked v1.4.1 aggregate. Scores and ranks can change when a new release is published; the page preview remains name-only.

JevBench v1.4.1 score

23.7

Rank #46 of 77 ranked systems.

39.6 points behind Jev 1.13.0's 63.3.

Where it sits among the 77 ranked systems. The marked tick is Jev 1.13.0 (63.3).

Published axes

intelligence
45.7
calibration
24.1
speed
86.3
cost
41.9

Accuracy per tier, incl. sealed

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: Jev 1.13.0 — Jev (TypeSafe, closed) · Score 63.3 (#1)
  • B: Raw Qwen3 8B direct logits — Raw-logit control (base model) · Score 23.7 (#46)

The four score axes

Radar: the four score axes, two systemsThe four score axes, Jev 1.13.0 vs Raw Qwen3 8B direct logits. Intelligence: 53.1 vs 45.7; Calibration: 76.3 vs 24.1; Speed: 83.3 vs 86.3; Cost: 52.0 vs 41.9.50100Intelligence53.1 · 45.7Calibration76.3 · 24.1Speed83.3 · 86.3Cost52.0 · 41.9
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, Jev 1.13.0 vs Raw Qwen3 8B direct logits. Easy: 100% vs 100%; Standard: 99% vs 86%; Judge: 95% vs 97%; Hard: 74% vs 46%; Sealed: 37% vs 26%.50100Easy100% · 100%Standard99% · 86%Judge95% · 97%Hard74% · 46%Sealed37% · 26%
Share correct per tier; Sealed = the 308 private decisions, aggregate only.

Hard tier by family (v1.2 topics)

Radar: hard tier by family (v1.2 topics), two systemsHard tier by family (v1.2 topics), Jev 1.13.0 vs Raw Qwen3 8B direct logits. Adversarial: 100% vs 58%; Ambiguous: 79% vs 43%; Judge: 79% vs 70%; Long policy: 61% vs 24%; Multi-hop: 86% vs 49%; Probability: 80% vs 40%; Routing: 100% vs 90%; Temporal / numeric: 27% vs 10%; Trade-off: 92% vs 42%; Trap: 100% vs 94%.50100Adversarial100% · 58%Ambiguous79% · 43%Judge79% · 70%Long policy61% · 24%Multi-hop86% · 49%Probability80% · 40%Routing100% · 90%Temporal /numeric27% · 10%Trade-off92% · 42%Trap100% · 94%
Share correct within each family of the 47 v1.2 hard-tier decisions (public and held-out).

Sealed set by family

Radar: sealed set by family, two systemsSealed set by family, Jev 1.13.0 vs Raw Qwen3 8B direct logits. Ambiguous / abstain: 30% vs 22%; Judge: 34% vs 37%; Long policy: 28% vs 15%; Multi-hop: 45% vs 21%; Paraphrase: 64% vs 57%; Probability: 50% vs 21%; Safety judge: 38% vs 31%; Temporal / numeric: 29% vs 27%; Trade-off: 38% vs 23%; Trap / adversarial: 42% vs 33%.50100Ambiguous /abstain30% · 22%Judge34% · 37%Long policy28% · 15%Multi-hop45% · 21%Paraphrase64% · 57%Probability50% · 21%Safety judge38% · 31%Temporal /numeric29% · 27%Trade-off38% · 23%Trap /adversarial42% · 33%
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
Apache-2.0
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our lium.io RTX A6000 48 GB; local in-process; serial; one forward pass
Published source
https://huggingface.co/Qwen/Qwen3-8B

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