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

decider-4b v2 (Mapika)

system-one-open · by Mapika

JevBench v1.4.2.2 score

64.1

Rank #3 of 91 ranked systems.

0.8 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
49.4
calibration
75.0
speed
92.9
cost
60.9

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, decider-4b v2 vs Jev 1.13.0. Intelligence: 49.4 vs 53.1; Calibration: 75.0 vs 76.3; Speed: 92.9 vs 83.3; Cost: 60.9 vs 52.0.50100Intelligence49.4 · 53.1Calibration75.0 · 76.3Speed92.9 · 83.3Cost60.9 · 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, decider-4b v2 vs Jev 1.13.0. Easy: 100% vs 100%; Standard: 97% vs 99%; Judge: 88% vs 95%; Hard: 67% vs 74%; Sealed: 35% vs 37%.50100Easy100% · 100%Standard97% · 99%Judge88% · 95%Hard67% · 74%Sealed35% · 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), decider-4b v2 vs Jev 1.13.0. Ambiguous / abstain: 51% vs 43%; Judge: 41% vs 54%; Long policy: 46% vs 44%; Multi-hop: 49% vs 64%; Probability: 54% vs 63%; Temporal / numeric: 34% vs 28%; Trade-off: 47% vs 55%; Routing: 100% vs 100%; Trap / adversarial: 78% vs 83%; Paraphrase: 50% vs 64%; Safety judge: 38% vs 38%.50100Ambiguous /abstain51% · 43%Judge41% · 54%Long policy46% · 44%Multi-hop49% · 64%Probability54% · 63%Temporal /numeric34% · 28%Trade-off47% · 55%Routing100% · 100%Trap /adversarial78% · 83%Paraphrase50% · 64%Safety judge38% · 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, decider-4b v2 vs Jev 1.13.0. Ambiguous / abstain: 43% vs 30%; Judge: 32% vs 34%; Long policy: 30% vs 28%; Multi-hop: 34% vs 45%; Paraphrase: 50% vs 64%; Probability: 36% vs 50%; Safety judge: 38% vs 38%; Temporal / numeric: 30% vs 29%; Trade-off: 35% vs 38%; Trap / adversarial: 33% vs 42%.50100Ambiguous /abstain43% · 30%Judge32% · 34%Long policy30% · 28%Multi-hop34% · 45%Paraphrase50% · 64%Probability36% · 50%Safety judge38% · 38%Temporal /numeric30% · 29%Trade-off35% · 38%Trap /adversarial33% · 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 (package and weights)
Cost evidence
estimated; the board’s row disclosure contains the published basis.
Endpoint condition
our evaluator-owned Lium GPU pod (RTX PRO 6000), offline read-only container, author's server on loopback
Note on this row
Decider-ai 1.2.2 (PyPI wheel identical to tag v1.2.2 abadc94), weights Mapika/decider-4b rev 7ab294cbdf6be6ac17fc818c10cdead744393d92 (decider_config version 4b-v2, T=1.935), uvicorn decider.serve:app. Disclosed by the author: 8,000 of the v2 LoRA rows come from generators written from the published names of the ten sealed families (no item read). Independent #1 gate (24 Sep): LEGIT. The author's private stage-2 training rows could not be audited for overlap with public items. Offline self-hosted inference of all 842 decisions (534 frozen v1.2 + 308 sealed v1.4) in a network-disabled, read-only container on an evaluator-owned Lium RTX PRO 6000 pod, through JevBench's unchanged typesafe adapter against the author's own server on loopback; no operator endpoint; no golds were exposed. Latency is the serial request wall time on the standard+judge items with the self-hosted adjustment. Cost is a labelled estimate: DeepInfra Qwen/Qwen3.5-4B list price $0.03/M input (4B dense size class, as decider-2b), read 2026-09-24, over the server's own usage.input_tokens; it is not a GPU bill.
Published source
https://github.com/Mapika/decider

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.