@gabepereyra: Harvey partnered with @appliedcompute to train a legal agent. We optimized each part of the agent stack, including the …
Summary
Harvey partnered with Applied Compute to train a legal agent, optimizing the agent stack and post-training the GLM-5.1 model using reward signals from their Legal Agent Benchmark.
View Cached Full Text
Cached at: 06/23/26, 03:51 PM
Harvey partnered with @appliedcompute to train a legal agent.
We optimized each part of the agent stack, including the eval loop, agent harness and compaction, and post-trained the underlying GLM-5.1 model using reward signal from Harvey’s Legal Agent Benchmark (LAB).
Check out more in the agent-training deep dive below.
Kudos to @nikogrupen, @ItsJulioPereyra, @rhythmrg, @jacob_dphillips, and @raymondmfeng for leading this effort - more to come, with lots of opportunity to push the frontier with GLM-5.2.
Similar Articles
Introducing Harvey's Legal Agent Benchmark (12 minute read)
Harvey releases the Legal Agent Benchmark (LAB), an open-source tool designed to evaluate AI agents on long-horizon legal tasks to help law firms measure ROI and track progress.
Customizing models for legal professionals
Harvey, a generative AI platform for legal professionals, partnered with OpenAI to create a custom-trained case law model that reduces hallucinations and improves reasoning for complex legal tasks like document drafting and contract analysis. The custom model, trained on 10 billion tokens of U.S. case law, achieved 97% lawyer preference over standard foundation models.
@ClementDelangue: Routing and post-training open-source models won't only give you more accurate systems but also meaningfully faster and…
Discussion on how routing and post-training open-source models can outperform frontier models in accuracy, speed, and cost, with Harvey's partnership with Fireworks AI demonstrating hybrid legal agents beating frontier models on quality and cost.
I built a multi-agent AI system for a mid-size law firm — here's what actually worked (and what didn't)
The author shares lessons learned from deploying a multi-agent AI system for a law firm using Claude and LangGraph, highlighting the success of confidence-score handoffs and the critical need for human-in-the-loop oversight to prevent hallucinations.
@aiDotEngineer: Your Agent Can Now Train Models The argument from @mervenoyann: open source models have caught up. GLM 5.1 is leading t…
The talk by @mervenoyann demonstrates that open source models like GLM 5.1 have caught up to closed models, and shows how Hugging Face's ecosystem enables agents to train models, run inference, and build workflows.