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

Winnow-12B Q8

Jev rebuild · by Eldan Ring

JevBench v1.4.2.2 score

55.6

Rank #8 of 91 ranked systems.

7.7 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
48.3
calibration
64.8
speed
82.3
cost
52.9

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.

The four score axes

Radar: the four score axes, two systemsThe four score axes, Winnow-12B Q8 vs Jev 1.13.0. Intelligence: 48.3 vs 53.1; Calibration: 64.8 vs 76.3; Speed: 82.3 vs 83.3; Cost: 52.9 vs 52.0.50100Intelligence48.3 · 53.1Calibration64.8 · 76.3Speed82.3 · 83.3Cost52.9 · 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, Winnow-12B Q8 vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 97% vs 99%; Judge: 91% vs 95%; Hard: 71% vs 74%; Sealed: 33% vs 37%.50100Easy100% · 100%Standard97% · 99%Judge91% · 95%Hard71% · 74%Sealed33% · 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), Winnow-12B Q8 vs Jev 1.13.0. Ambiguous / abstain: 49% vs 43%; Judge: 65% vs 54%; Long policy: 44% vs 44%; Multi-hop: 52% vs 64%; Probability: 42% vs 63%; Temporal / numeric: 20% vs 28%; Trade-off: 42% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 83% vs 83%; Paraphrase: 64% vs 64%; Safety judge: 50% vs 38%.50100Ambiguous /abstain49% · 43%Judge65% · 54%Long policy44% · 44%Multi-hop52% · 64%Probability42% · 63%Temporal /numeric20% · 28%Trade-off42% · 55%Routing100% · 100%Trap /adversarial83% · 83%Paraphrase64% · 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, Winnow-12B Q8 vs Jev 1.13.0. Ambiguous / abstain: 35% vs 30%; Judge: 46% vs 34%; Long policy: 23% vs 28%; Multi-hop: 37% vs 45%; Paraphrase: 64% vs 64%; Probability: 32% vs 50%; Safety judge: 50% vs 38%; Temporal / numeric: 16% vs 29%; Trade-off: 23% vs 38%; Trap / adversarial: 50% vs 42%.50100Ambiguous /abstain35% · 30%Judge46% · 34%Long policy23% · 28%Multi-hop37% · 45%Paraphrase64% · 64%Probability32% · 50%Safety judge50% · 38%Temporal /numeric16% · 29%Trade-off23% · 38%Trap /adversarial50% · 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, including the applicable Gemma 4 base/derivative licence terms
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our GPU (lium.io RTX 4090 24 GB), reached over the internet from Germany; serial, one request at a time
Note on this row
The submitted Q8_0 GGUF ran through the pinned author's TypeSafe-compatible /v1/systemone server with 8,192 context, four resident decision branches, Q8 KV, and full GPU offload. The private training corpus was not released. The author's checksum-based audit reports zero exact public-item overlap, but that claim cannot be independently reproduced; our scan found no exact public state or instruction text in the released artifacts. Cost uses the $0.05/M-input hosted Gemma 3 12B reference, not free/100.
Published source
https://huggingface.co/EldanRing/Winnow-12B

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.