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

reflex 4B (kshetrajna12)

Jev rebuild · by kshetrajna12

JevBench v1.4.2.2 score

54.0

Rank #9 of 91 ranked systems.

9.3 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
47.5
calibration
70.4
speed
68.0
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: reflex 4B — Jev rebuild · Score 54.0 (#9)
  • 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, reflex 4B vs Jev 1.13.0. Intelligence: 47.5 vs 53.1; Calibration: 70.4 vs 76.3; Speed: 68.0 vs 83.3; Cost: 59.7 vs 52.0.50100Intelligence47.5 · 53.1Calibration70.4 · 76.3Speed68.0 · 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, reflex 4B vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 95% vs 99%; Judge: 97% vs 95%; Hard: 63% vs 74%; Sealed: 28% vs 37%.50100Easy100% · 100%Standard95% · 99%Judge97% · 95%Hard63% · 74%Sealed28% · 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), reflex 4B vs Jev 1.13.0. Ambiguous / abstain: 31% vs 43%; Judge: 51% vs 54%; Long policy: 36% vs 44%; Multi-hop: 48% vs 64%; Probability: 40% vs 63%; Temporal / numeric: 27% vs 28%; Trade-off: 42% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 78% vs 83%; Paraphrase: 21% vs 64%; Safety judge: 44% vs 38%.50100Ambiguous /abstain31% · 43%Judge51% · 54%Long policy36% · 44%Multi-hop48% · 64%Probability40% · 63%Temporal /numeric27% · 28%Trade-off42% · 55%Routing100% · 100%Trap /adversarial78% · 83%Paraphrase21% · 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, reflex 4B vs Jev 1.13.0. Ambiguous / abstain: 24% vs 30%; Judge: 34% vs 34%; Long policy: 20% vs 28%; Multi-hop: 34% vs 45%; Paraphrase: 21% vs 64%; Probability: 29% vs 50%; Safety judge: 44% vs 38%; Temporal / numeric: 23% vs 29%; Trade-off: 27% vs 38%; Trap / adversarial: 42% vs 42%.50100Ambiguous /abstain24% · 30%Judge34% · 34%Long policy20% · 28%Multi-hop34% · 45%Paraphrase21% · 64%Probability29% · 50%Safety judge44% · 38%Temporal /numeric23% · 29%Trade-off27% · 38%Trap /adversarial42% · 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
MIT (code, adapter); Apache-2.0 (base)
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our RunPod GPU (H100 NVL 96 GB, Canada), reached over the internet from Germany
Note on this row
The author's reflex-serve: Qwen3.5-4B with the published LoRA and its per-primitive calibration file; the state is encoded once and each question read from the label logits. Run serially on our GPU; the author discloses that the 231 public items were used four times as a development gate.
Published source
https://github.com/kshetrajna12/reflex

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.