Introducing amicus-embed-ng-v1: The Sovereign Legal Embedding Model for Commonwealth Jurisdictions

General embedding models (such as OpenAI's text-embedding-3-small or Cohere's English models) often miss deep judicial nuances, landmark precedent structures, and legislative context specific to Commonwealth jurisdictions.

amicus-embed-ng-v1 is our answer. Fine-tuned specifically on millions of expert-curated Commonwealth legal query-precedent pairs, authority citations, and statutory acts, it represents a hyper-accurate, high-recall retriever optimized for complex litigation advocacy globally.

Improving on standard model bounds, amicus-embed-ng-v1 features a context window of 8,192 tokens. This allows developers to embed entire statutory chapters, complex contracts, or multi-page judicial briefs in a single API request without loss of context due to chunking.

Additionally, the model is fully Matryoshka-aware, supporting truncation down to 768, 512, or 256 dimensions. At 768 dimensions, it provides optimal authority alignment and dense legal retrieval capability with extremely low storage requirements.

On the Model Legal Embedding Benchmark (MLEB) slice for Commonwealth precedents, amicus-embed-ng-v1 establishes a new state-of-the-art frontier:

• Legal Search Accuracy (NDCG@10): 94.8% (Outperforming OpenAI's best models by over 23%).
• Citation Recall (Recall@5): 92.3%, ensuring that AI agents can find landmark Supreme Court citations with near-perfect alignment.
• Statutory Mapping: 88.6% precision mapping factual pattern queries directly to criminal and penal code acts.

The model is now accessible to the public via the Dockase Developer Gateway. Billed at an extremely competitive rate of $0.25 per 1 Million tokens, it offers developers a cost-effective, high-fidelity domain-adapted alternative to generic commercial model keys.

We've also automated overage billing using Paystack card tokenization, guaranteeing that developer workloads are never interrupted mid-execution.