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

OpenSourceJev (Qwen3.5-4B Q4_K_M, native llama.cpp)

Jev rebuild · by sabeel111

JevBench v1.4.2.2 score

40.9

Rank #23 of 91 ranked systems.

22.4 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
41.8
calibration
60.3
speed
82.0
cost
64.0

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.

The four score axes

Radar: the four score axes, two systemsThe four score axes, OpenSourceJev vs Jev 1.13.0. Intelligence: 41.8 vs 53.1; Calibration: 60.3 vs 76.3; Speed: 82.0 vs 83.3; Cost: 64.0 vs 52.0.50100Intelligence41.8 · 53.1Calibration60.3 · 76.3Speed82.0 · 83.3Cost64.0 · 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, OpenSourceJev vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 93% vs 99%; Judge: 86% vs 95%; Hard: 56% vs 74%; Sealed: 26% vs 37%.50100Easy100% · 100%Standard93% · 99%Judge86% · 95%Hard56% · 74%Sealed26% · 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), OpenSourceJev vs Jev 1.13.0. Ambiguous / abstain: 33% vs 43%; Judge: 53% vs 54%; Long policy: 24% vs 44%; Multi-hop: 40% vs 64%; Probability: 35% vs 63%; Temporal / numeric: 20% vs 28%; Trade-off: 39% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 68% vs 83%; Paraphrase: 50% vs 64%; Safety judge: 50% vs 38%.50100Ambiguous /abstain33% · 43%Judge53% · 54%Long policy24% · 44%Multi-hop40% · 64%Probability35% · 63%Temporal /numeric20% · 28%Trade-off39% · 55%Routing100% · 100%Trap /adversarial68% · 83%Paraphrase50% · 64%Safety judge50% · 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, OpenSourceJev vs Jev 1.13.0. Ambiguous / abstain: 22% vs 30%; Judge: 39% vs 34%; Long policy: 10% vs 28%; Multi-hop: 24% vs 45%; Paraphrase: 50% vs 64%; Probability: 32% vs 50%; Safety judge: 50% vs 38%; Temporal / numeric: 16% vs 29%; Trade-off: 31% vs 38%; Trap / adversarial: 25% vs 42%.50100Ambiguous /abstain22% · 30%Judge39% · 34%Long policy10% · 28%Multi-hop24% · 45%Paraphrase50% · 64%Probability32% · 50%Safety judge50% · 38%Temporal /numeric16% · 29%Trade-off31% · 38%Trap /adversarial25% · 42%
Share correct within each sealed family — system-level aggregates; the items stay private.

Availability and evidence

Openness
Unknown in the published row
License note
MIT (repository code); Apache-2.0 (Qwen/Qwen3.5-4B base and unsloth/Qwen3.5-4B-GGUF Q4_K_M conversion)
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
re-run on a throwaway RunPod pod with the original recipe; deviations in its manifest
Note on this row
DM submission, measure-only (JevBench publishing HOLD in force). Same author as the existing simplejev-qwen3.5-0.8b row (sabeel111/Featherless AI). Round-4 audit of an earlier commit could not be measured (no public GGUF, Windows-only DLL loader, unpinned llama.cpp build); this round the author published the exact unsloth Q4_K_M GGUF (hash/size independently verified) and we built llama.cpp CUDA from current upstream master on Linux ourselves -- its ABI matched the ctypes bindings exactly, so only a loader file-naming fix was needed (documented diff), no code/scoring/calibration change. Our public-231 subset exactly reproduced the author-reported table: easy 48/48, standard 67/72, hard 66/111, schema 231/231. Calibration (noul temperature) fit only on Google BoolQ, not JevBench. Repo docs name 3 public task IDs while describing benchmark-directed algorithm fixes on the public half (disclosed); 0 exact state/instruction text matches in a released-file scan.
Published source
https://github.com/sabeel111/OpenSourceJev

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.