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

Imajev-4B

Jev rebuild · by mohit67890

JevBench v1.4.2.2 score

67.4

Rank #1 of 91 ranked systems.

4.1 points ahead of 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
52.2
calibration
80.4
speed
90.6
cost
59.7

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: Imajev-4B — Jev rebuild · Score 67.4 (#1)
  • 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, Imajev-4B vs Jev 1.13.0. Intelligence: 52.2 vs 53.1; Calibration: 80.4 vs 76.3; Speed: 90.6 vs 83.3; Cost: 59.7 vs 52.0.50100Intelligence52.2 · 53.1Calibration80.4 · 76.3Speed90.6 · 83.3Cost59.7 · 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, Imajev-4B vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 99% vs 99%; Judge: 89% vs 95%; Hard: — vs 74%; Sealed: 37% vs 37%.50100Easy100% · 100%Standard99% · 99%Judge89% · 95%Hard74%Sealed37% · 37%
Share correct per tier; Sealed = the 308 private decisions, aggregate only.

Current question set by family (hard + sealed)

Imajev-4B has no published hard-tier family breakdown; families that need it are left out (—).

Radar: current question set by family (hard + sealed), two systemsCurrent question set by family (hard + sealed), Imajev-4B vs Jev 1.13.0. Ambiguous / abstain: — vs 43%; Judge: — vs 54%; Long policy: — vs 44%; Multi-hop: — vs 64%; Probability: — vs 63%; Temporal / numeric: — vs 28%; Trade-off: — vs 55%; Routing: — vs 100%; Trap / adversarial: — vs 83%; Paraphrase: 50% vs 64%; Safety judge: 44% vs 38%.50100Ambiguous /abstain43%Judge54%Long policy44%Multi-hop64%Probability63%Temporal /numeric28%Trade-off55%Routing100%Trap /adversarial83%Paraphrase50% · 64%Safety judge44% · 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, Imajev-4B vs Jev 1.13.0. Ambiguous / abstain: 41% vs 30%; Judge: 44% vs 34%; Long policy: 33% vs 28%; Multi-hop: 34% vs 45%; Paraphrase: 50% vs 64%; Probability: 25% vs 50%; Safety judge: 44% vs 38%; Temporal / numeric: 30% vs 29%; Trade-off: 38% vs 38%; Trap / adversarial: 58% vs 42%.50100Ambiguous /abstain41% · 30%Judge44% · 34%Long policy33% · 28%Multi-hop34% · 45%Paraphrase50% · 64%Probability25% · 50%Safety judge44% · 38%Temporal /numeric30% · 29%Trade-off38% · 38%Trap /adversarial58% · 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
Apache-2.0 (author adapter and server metadata)
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
Evaluator-owned Lium GPU pod; network-disabled, read-only container; the author's reviewed server ran on loopback.
Note on this row
Imajev-4B: requested adapter c9e5f132465da85d31735ec502d5557982671a7d and server a0134749e0900189c129cd6bb5000969f3b64bb5; one rotation with calibration.json. The optimized image included flash-linear-attention/fla-core 0.5.2 and causal-conv1d 1.7.0; CUDA profiling confirmed the pinned kernels ran. Cost is an estimate using the public DeepInfra Qwen/Qwen3.5-4B base-model reference price ($0.03/M input, $0.15/M output), with no generated output tokens; it is not a GPU bill.
Published source
https://github.com/mohit67890/imajev

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.