Mapika Decider 4B Release - JevBench Standard: leaderboard

Mapika Decider 4B Release - JevBench Standard: Accuracy fraction (0-1). Mapika decider-4b official v3 card: trained Qwen3.5-4B-Base decision model, compared with released older v2.1. Model update is distinct trained weights, not an inference setup. DI 0.3 4500-request sample and JevBench public hard/standard accuracy retained; image-mean old result adds base vision tower, so not presented as the same checkpoint condition. Fresh held-out JevBench-style skill index can be negative: explicit unbounded theoretical scale, not a percentage.

Metric: Accuracy fraction (0-1). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 2 models tracked.

Top models

Interactive version: theaggregate.ai/benchmark?slug=mapika-decider-4b-release-jevbench-standard · How It Works · Data refreshed daily, snapshot 2026-11-12.