Contrastive-LM Releases CLM-8B: An Open System One Model That Scores Agent Actions Up to 9× Faster Than Jev
9×. That's how much faster Contrastive-LM's new open model scores agent actions compared to the current baseline — and it's built on a frozen encoder, not a generative model.
Why it matters
A specialized 'System One' architecture for agent decision-scoring is emerging as a faster, more efficient alternative to using full generative models for agentic workflows. CLM-8B's speed and open weights could reshape how practitioners build agent verifiers and action rankers at scale.
The key facts
7 to knowCLM-8B: 8B-parameter open model built on frozen Qwen3-8B encoder
Architecture: 2 small projection heads trained with contrastive InfoNCE objective
Speed: up to 9× faster than TypeSafe Jev in zero-shot action scoring
Accuracy: 81.6% on held-out DeepSWE tasks; 87.6% on held-out Terminal-Bench 2.1
Use case: scoring candidate actions against state (not text generation)
Released by: Contrastive-LM
Status: open weights available
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Contrastive-LM has released CLM-8B, an open System One model that scores candidate actions against a state instead of generating text. It adds 2 small projection heads to a frozen Qwen3-8B encoder and trains them with a contrastive InfoNCE objective. In zero-shot tests it runs up to 9× faster than…