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

djev (thinking)

Jev rebuild · by David Villalon / Maisa

JevBench v1.4.2.2 score

15.2

Rank #68 of 91 ranked systems.

48.1 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
71.6
calibration
87.8
speed
75.2
cost
26.9

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: djev — Jev rebuild · Score 15.2 (#68)
  • 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, djev vs Jev 1.13.0. Intelligence: 71.6 vs 53.1; Calibration: 87.8 vs 76.3; Speed: 75.2 vs 83.3; Cost: 26.9 vs 52.0.50100Intelligence71.6 · 53.1Calibration87.8 · 76.3Speed75.2 · 83.3Cost26.9 · 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, djev vs Jev 1.13.0. Easy: 96% vs 100%; Standard: 99% vs 99%; Judge: 80% vs 95%; Hard: 78% vs 74%; Sealed: 60% vs 37%.50100Easy96% · 100%Standard99% · 99%Judge80% · 95%Hard78% · 74%Sealed60% · 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), djev vs Jev 1.13.0. Ambiguous / abstain: 51% vs 43%; Judge: 77% vs 54%; Long policy: 56% vs 44%; Multi-hop: 66% vs 64%; Probability: 85% vs 63%; Temporal / numeric: 64% vs 28%; Trade-off: 53% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 88% vs 83%; Paraphrase: 79% vs 64%; Safety judge: 56% vs 38%.50100Ambiguous /abstain51% · 43%Judge77% · 54%Long policy56% · 44%Multi-hop66% · 64%Probability85% · 63%Temporal /numeric64% · 28%Trade-off53% · 55%Routing100% · 100%Trap /adversarial88% · 83%Paraphrase79% · 64%Safety judge56% · 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, djev vs Jev 1.13.0. Ambiguous / abstain: 51% vs 30%; Judge: 66% vs 34%; Long policy: 53% vs 28%; Multi-hop: 58% vs 45%; Paraphrase: 79% vs 64%; Probability: 79% vs 50%; Safety judge: 56% vs 38%; Temporal / numeric: 61% vs 29%; Trade-off: 38% vs 38%; Trap / adversarial: 83% vs 42%.50100Ambiguous /abstain51% · 30%Judge66% · 34%Long policy53% · 28%Multi-hop58% · 45%Paraphrase79% · 64%Probability79% · 50%Safety judge56% · 38%Temporal /numeric61% · 29%Trade-off38% · 38%Trap /adversarial83% · 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
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our GPU (lium.io H200 141 GB), reached over the internet from Germany; serial, one request at a time
Note on this row
Experimental full-generation path over the same DiffusionGemma checkpoint as djev-dev: thinking was enabled and the model could generate up to 8,192 tokens before returning its distribution. Current djev-dev itself hard-codes enable_thinking=false, diffusion_max_steps=1 and read_only=true, so this is not a switch in its published typed API. It is substantially slower/costlier, and 72/534 requests exhausted the output budget without a parseable distribution; those are failures. Cost uses measured tokens and a same-size hosted reference, not the H200 rental bill. 220/308 sealed items answered validly (failures count as wrong)
Published source
https://github.com/Davipar/djev-dev

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.