---
title: "Discussion on \"Decision models against small LLMs: Jev, Kev, Laya and Mammoth-1 on 6,662 decisions\" | Hashnode"
description: "Discussion on \"Decision models against small LLMs: Jev, Kev, Laya and Mammoth-1 on 6,662 decisions\". Four decision models and five small LLMs, tested on six public datasets with the harness and raw calls published.\nhttps://pournasserian.com/writing/decision-models-vs-small-llms-2026"
url: "https://hashnode.com/posts/decision-models-against-small-llms-jev-kev-laya-and-mammoth-1-on-6-662-decisions/6aed130b48adf958e957b027"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Decision models against small LLMs: Jev, Kev, Laya and Mammoth-1 on 6,662 decisions](https://pournasserian.hashnode.dev/decision-models-against-small-llms-jev-kev-laya-and-mammoth-1-on-6-662-decisions?utm_source=hashnode&utm_medium=feed)

Four decision models and five small LLMs, tested on six public datasets with the harness and raw calls published. https://pournasserian.com/writing/decision-models-vs-small-llms-2026

Nov 5 · 1 min read · pournasserian.hashnode.dev  
Read the article

-   [#ai](https://hashnode.com/tag/ai)
-   [#llm](https://hashnode.com/tag/llm)
-   [#aiagents](https://hashnode.com/tag/aiagents)

## Written by

[Amir Pournasserian](https://hashnode.com/@pournasserian)

@pournasserian · Toronto, Canada

  
[View profile](https://hashnode.com/@pournasserian)

## Responses 1

Markdown supported · ⌘↵ to post

[Ahmet Özel](https://hashnode.com/@ahmetozel)

[6d ago](https://hashnode.com/posts/decision-models-against-small-llms-jev-kev-laya-and-mammoth-1-on-6-662-decisions/6aed130b48adf958e957b027/comment/6aed9623d889f2b52f41e098)

The CLINC escape-hatch table makes a deployment decision much clearer than the overall mean. Mammoth-1 catching 152 of 190 out-of-scope requests with 100 false alarms and Kev catching 136 with 239 false alarms are materially different operating points, even before latency enters the choice.

A useful next view would separate false in-scope routing from unnecessary rejection across probability thresholds, using a separate calibration split. That would show which models offer a usable tradeoff when a wrong destination costs more than sending a request for review, without assuming the default argmax is each model's best operating point.