<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/kev-is-a-family-of-four-open-source-decision-models-ranging-from-0-8b-to-27b-parameters-it-uses-the-se77r60oa" -->

---
title: Kev is a family of four open-source decision models...
description: A new release of Kev, an open source family of four decision models (0.8B to 27B parameters) under Apache-2.0 license, designed to take text, questions, and...
canonical: https://daily.dev/posts/kev-is-a-family-of-four-open-source-decision-models-ranging-from-0-8b-to-27b-parameters-it-uses-the-se77r60oa
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Kev is a family of four open-source decision models ranging from 0.8B to 27B parameters. It uses the same API as TypeSafe's Jev, so applications built with TypeSafe's SDK can switch just by changing the endpoint and model name. | daily.dev
og:description: A new release of Kev, an open source family of four decision models (0.8B to 27B parameters) under Apache-2.0 license, designed to take text, questions, and...
og:url: https://daily.dev/posts/kev-is-a-family-of-four-open-source-decision-models-ranging-from-0-8b-to-27b-parameters-it-uses-the-se77r60oa
og:image: https://api.daily.dev/og/posts/SE77R60OA.png
og:image:alt: Kev is a family of four open-source decision models ranging from 0.8B to 27B parameters. It uses the same API as TypeSafe's Jev, so applications built with TypeSafe's SDK can switch just by changing the endpoint and model name.
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Kev is a family of four open-source decision models ranging from 0.8B to 27B parameters. It uses the same API as TypeSafe's Jev, so applications built with TypeSafe's SDK can switch just by changing the endpoint and model name.

**[Andrew M](https://daily.dev/sources/dup6ye036hupdshygexyu)** · [@andrewma](https://daily.dev/andrewma) · 50 upvotes · 5 comments

## Summary

A new release of Kev, an open source family of four decision models (0.8B to 27B parameters) under Apache-2.0 license, designed to take text, questions, and candidate answers and return calibrated probabilities rather than generated text. Kev-27B is newly added, fine-tuned from Qwen3.8-27B across the entire backbone, while smaller models use LoRA adapters on Qwen3.5 backbones. The models expose an API compatible with TypeSafe's Jev, run on GPUs or Apple Silicon via MLX, and come with evaluation results across six panels covering generalization, short-text decisions, policy reasoning, consumer complaints, and developer-tool decisions. Detailed benchmarks cover accuracy, calibration error, confidence thresholds, caching performance for long documents, training cost (~$650 for a full 27B run, ~$1 for small fine-tunes), and inference latency/cost across GPU tiers. Coding-agent skills (kev-deploy, kev-finetune) are provided to deploy and fine-tune models directly from training data or tickets.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://jaredpalmer.com/blog/introducing-kev>

## Community discussion

Top comments from developers on daily.dev.

**@pdfopsdev** · 3 upvotes

> does it match jev's fallback/routing behavior when a model's unavailable, or just the request shape? drop-in api compat usually breaks on the edge cases, not the happy path.

**@ahmetozel** · 1 upvotes

> The serving caveats matter as much as the endpoint compatibility. The reported model-time medians exclude network latency and cold starts, while the scale-to-zero setup has a measured cold start of about thirty-five seconds. Those are very different user-visible regimes for a short decision request.
>
> I would benchmark the actual arrival pattern: a cold first request, warm requests, and concurrent requests sharing or replacing the document prefix. Report end-to-end latency and cache residency alongside the model time. That would help choose among the smaller models without assuming a...

**@lazyown** · 1 upvotes

> i will try :D

**@smoky1496** · 1 upvotes

> **I'll give this repo a try and see how it works in practice**

**@debadrita\_eduonix** · 0 upvotes

> **AI agents need efficient engineering, not just capable models.**
>
> As AI-powered applications become more common, inference speed, computational costs, and practical implementation matter just as much as model capabilities.
>
> The Complete Jev Application Engineering Masterclass is a Kickstarter project exploring AI agent development and efficiency in classification tasks.
>
> Curious about the approach? Here's the project:
>
> https://www.kickstarter.com/projects/eduonix/complete-jev-application-engineering-masterclass?ref=a7oiay

---

Tags: [#open-source](https://daily.dev/tags/open-source), [#deep-learning](https://daily.dev/tags/deep-learning), [#qwen](https://daily.dev/tags/qwen)

[View this post on daily.dev](https://daily.dev/posts/kev-is-a-family-of-four-open-source-decision-models-ranging-from-0-8b-to-27b-parameters-it-uses-the-se77r60oa)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/kev-is-a-family-of-four-open-source-decision-models-ranging-from-0-8b-to-27b-parameters-it-uses-the-se77r60oa","headline":"Kev is a family of four open-source decision models ranging from 0.8B to 27B parameters. It uses the same API as TypeSafe's Jev, so applications built with TypeSafe's SDK can switch just by changing the endpoint and model name.","text":"Shared: Introducing Kev","url":"https://daily.dev/posts/kev-is-a-family-of-four-open-source-decision-models-ranging-from-0-8b-to-27b-parameters-it-uses-the-se77r60oa","datePublished":"2026-10-02T23:10:08.637Z","dateModified":"2026-10-02T23:11:17.706Z","author":{"@type":"Person","name":"Andrew M","url":"https://daily.dev/andrewma","image":"https://avatars.githubusercontent.com/u/102819214?v=4","description":"full time overthinker\n","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":3760}},"image":"https://media.daily.dev/image/upload/s--qPvKM23u--/f_auto/v1722860399/public/Placeholder%2009","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":50},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":5}],"sharedContent":{"@type":"WebPage","url":"https://api.daily.dev/r/8Olh6x7Jr"},"comment":[{"@type":"Comment","text":"does it match jev’s fallback/routing behavior when a model’s unavailable, or just the request shape? drop-in api compat usually breaks on the edge cases, not the happy path.","datePublished":"2026-10-03T18:05:15.464Z","url":"https://daily.dev/posts/SE77R60OA#c-C94vWk0Ip","author":{"@type":"Person","name":"PDFops","url":"https://daily.dev/pdfopsdev","image":"https://media.daily.dev/image/upload/s---8isRBKc--/f_auto/v1782922291/avatars/avatar_orjMeK8QKaaVZwGq7ScPz?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":3}},{"@type":"Comment","text":"The serving caveats matter as much as the endpoint compatibility. The reported model-time medians exclude network latency and cold starts, while the scale-to-zero setup has a measured cold start of about thirty-five seconds. Those are very different user-visible regimes for a short decision request.\nI would benchmark the actual arrival pattern: a cold first request, warm requests, and concurrent requests sharing or replacing the document prefix. Report end-to-end latency and cache residency alongside the model time. That would help choose among the smaller models without assuming a continuously busy GPU or a permanently warm document cache.","datePublished":"2026-10-07T14:36:40.310Z","url":"https://daily.dev/posts/SE77R60OA#c-iLaW7SeyB","author":{"@type":"Person","name":"Ahmet Özel","url":"https://daily.dev/ahmetozel","image":"https://avatars.githubusercontent.com/u/70992231?v=4"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}},{"@type":"Comment","text":"i will try :D","datePublished":"2026-10-08T01:29:59.483Z","url":"https://daily.dev/posts/SE77R60OA#c-C5SEBe3Hp","author":{"@type":"Person","name":"Lazy Own","url":"https://daily.dev/lazyown","image":"https://lh3.googleusercontent.com/a/ACg8ocI_o0Fl9YqEsGGkLg5IdLmkmUzB0UsXpVq9L_K4JqauWH1rhw=s96-c"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}},{"@type":"Comment","text":"I’ll give this repo a try and see how it works in practice","datePublished":"2026-10-08T17:18:13.368Z","url":"https://daily.dev/posts/SE77R60OA#c-w7QnHBWON","author":{"@type":"Person","name":"Lượng Lê","url":"https://daily.dev/smoky1496","image":"https://media.daily.dev/image/upload/s--LsBXrQbv--/f_auto/v1791387179/avatars/avatar_ZKLbQ1ROTXM8EJ7PZv383?_a=BAMAMicg0"},"interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1}},{"@type":"Comment","text":"AI agents need efficient engineering, not just capable models.\nAs AI-powered applications become more common, inference speed, computational costs, and practical implementation matter just as much as model capabilities.\nThe Complete Jev Application Engineering Masterclass is a Kickstarter project exploring AI agent development and efficiency in classification tasks.\nCurious about the approach? Here’s the project:\nhttps://www.kickstarter.com/projects/eduonix/complete-jev-application-engineering-masterclass?ref=a7oiay","datePublished":"2026-10-29T10:03:18.141Z","url":"https://daily.dev/posts/SE77R60OA#c-BzDJcRZrT","author":{"@type":"Person","name":"Debadrita_eduonix","url":"https://daily.dev/debadrita_eduonix","image":"https://lh3.googleusercontent.com/a/ACg8ocIU5-gwv62zEE7TYZ-JOmLu_T15N_H7ZDYlBXqHb-unYFuBiw=s96-c"}}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/dup6ye036hupdshygexyu","name":"Andrew M"}}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Andrew M","item":"https://daily.dev/sources/dup6ye036hupdshygexyu"},{"@type":"ListItem","position":3,"name":"Kev is a family of four open-source decision models ranging from 0.8B to 27B parameters. It uses the same API as TypeSafe's Jev, so applications built with TypeSafe's SDK can switch just by changing the endpoint and model name."}]}
```