---
title: "Discussion on \"Fine-Tuning a 32B Legal LLM That Outperformed a Frontier Model at 4× Lower Serving Cost\" | Hashnode"
description: "Discussion on \"Fine-Tuning a 32B Legal LLM That Outperformed a Frontier Model at 4× Lower Serving Cost\". LegML fine-tuned a 32B legal LLM using GRPO and supervised fine-tuning that beat a frontier model on legal benchmarks."
url: "https://hashnode.com/posts/fine-tuning-32b-legal-llm-outperformed-frontier-model-lower-cost/69e782f1feb2629cd50851d2"
---

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

Discussion

# [Fine-Tuning a 32B Legal LLM That Outperformed a Frontier Model at 4× Lower Serving Cost](https://flexai.hashnode.dev/fine-tuning-32b-legal-llm-outperformed-frontier-model-lower-cost?utm_source=hashnode&utm_medium=feed)

LegML fine-tuned a 32B legal LLM using GRPO and supervised fine-tuning that beat a frontier model on legal benchmarks.

Apr 21 · 6 min read · flexai.hashnode.dev  
Read the article

-   [#flexai-news](https://hashnode.com/tag/flexai-news)
-   [#llm](https://hashnode.com/tag/llm)
-   [#fine-tuning-models](https://hashnode.com/tag/fine-tuning-models)
-   [#legal-ai](https://hashnode.com/tag/legal-ai)
-   [#sovereign-ai](https://hashnode.com/tag/sovereign-ai)
-   [#gpu](https://hashnode.com/tag/gpu)
-   [#h100](https://hashnode.com/tag/h100)
-   [#inference](https://hashnode.com/tag/inference)

## Written by

[Adam King](https://hashnode.com/@adamatflex)

@adamatflex · Austin, TX

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

BizOps at FlexAI

## Responses

Markdown supported · ⌘↵ to post

No responses yet.