AnyJev
Typed decisions with real probabilities from LLMs: the training-free L0 readout and the self-distilled Tacit models.
Updated • 1Note Training-free: AnyJev L0 on the unchanged google/gemma-4-31B-it, no weights of its own. Decision Index 0.3 Full score 60.22, 5th of 149 entries (2026-10-12).
morriszjm/Tacit-9B
Image-Text-to-Text • 9B • Updated • 432 • 2Note Self-distilled from Qwen3.5-9B. One forward pass per decision.
morriszjm/Tacit-8B
Text Generation • 8B • Updated • 1.09kNote Self-distilled from Qwen3-8B. One forward pass per decision.
morriszjm/Tacit-4B
Text Generation • 4B • Updated • 1.23kNote Self-distilled from Qwen3-4B. One forward pass per decision.
morriszjm/Tacit-2B
Image-Text-to-Text • 2B • Updated • 255Note Self-distilled from Qwen3.5-2B. One forward pass per decision.
morriszjm/Tacit-1.7B
Text Generation • 2B • Updated • 2.12kNote Self-distilled from Qwen3-1.7B. One forward pass per decision.
morriszjm/decision-index-results
Updated • 59Note The complete public Decision Index 0.3 run of AnyJev L0 · Gemma-4-31B.
Jev Decision Index
🔬720Benchmarks and news on various repros of TypeSafe's Jev
Note The Decision Index board.