Sold by: Isaacus
Deployed on AWS
A state-of-the-art legal embedding model optimized for semantic search, RAG, and document classification, ranked first on the Massive Legal Embedding Benchmark.
Overview
Kanon 2 Embedder is a state-of-the-art legal embedding model optimized for semantic search, RAG, and document classification. It achieves the highest information retrieval performance on the Massive Legal Embedding Benchmark (MLEB) out of 20 other models, including OpenAI Text Embedding 3 Large, Gemini Embedding, Voyage 3 Large, Qwen 3 Embedding 8B, and Jina Embeddings v4 (https://arxiv.org/abs/2510.19365 ). It also ranks first on case, legislation, and regulation retrieval and third on contract retrieval.
Kanon 2 Embedder is Matryoshka-aware, supporting truncation at 1,024, 768, 512, and 256 dimensions, with its default and maximum dimensionality being 1,792. Even at 256 dimensions, it manages to rank third on MLEB.
Kanon 2 Embedder has a context window of up to 16,384 tokens, allowing it to embed especially long legal documents.
On a g6e.xlarge instance, it can embed up to 1 billion tokens (roughly 250k average-length legal documents) per hour.
Like all other Isaacus SageMaker model deployments, your Kanon 2 Embedder deployment will be fully air-gapped--no data will enter or leave your AWS account.
Conveniently, Isaacus SageMaker model deployments are also compatible with the standard Isaacus Python SDK via the Isaacus SageMaker Python integration (https://docs.isaacus.com/integrations/amazon-sagemaker ).
You can negotiate a discount to Kanon 2 Embedder by contacting us: https://isaacus.com/support .
Kanon 2 Embedder can also be purchased in a discounted bundle alongside Kanon Universal Classifier here: https://aws.amazon.com/marketplace/pp/prodview-lquokmsovgpsm .
Highlights
- Ranked first on the Massive Legal Embedding Benchmark (MLEB) (https://huggingface.co/papers/2510.19365) at legal retrieval out of 20 other models, including OpenAI Text Embedding 3 Large, Gemini Embedding, Voyage 3 Large, Qwen 3 Embedding 8B, and Jina Embeddings v4.
- Supports Matryoshka truncation of embeddings down to 256 dimensions while still outperforming OpenAI Text Embedding 3 Large and Gemini Embedding on MLEB at full dimensionality.
- Capable of embedding texts up to 16,384 tokens in length, or roughly 46 pages of an average legal document, with throughput of 1 billion tokens (~250k documents) per hour on a g6e.xlarge instance.
Details
Sold by
Delivery method
Latest version
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
Pricing
Pricing is based on actual usage, with charges varying according to how much you consume. Subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimension | Description | Cost/host/hour |
|---|---|---|
ml.g6.xlarge Inference (Batch) Recommended | Model inference on the ml.g6.xlarge instance type, batch mode | $4.99 |
ml.g6e.xlarge Inference (Real-Time) Recommended | Model inference on the ml.g6e.xlarge instance type, real-time mode | $4.99 |
ml.g5.2xlarge Inference (Batch) | Model inference on the ml.g5.2xlarge instance type, batch mode | $4.99 |
ml.g5.4xlarge Inference (Batch) | Model inference on the ml.g5.4xlarge instance type, batch mode | $4.99 |
ml.g5.8xlarge Inference (Batch) | Model inference on the ml.g5.8xlarge instance type, batch mode | $4.99 |
ml.g5.16xlarge Inference (Batch) | Model inference on the ml.g5.16xlarge instance type, batch mode | $4.99 |
ml.g6.2xlarge Inference (Batch) | Model inference on the ml.g6.2xlarge instance type, batch mode | $4.99 |
ml.g6.4xlarge Inference (Batch) | Model inference on the ml.g6.4xlarge instance type, batch mode | $4.99 |
ml.g6.8xlarge Inference (Batch) | Model inference on the ml.g6.8xlarge instance type, batch mode | $4.99 |
ml.g6.16xlarge Inference (Batch) | Model inference on the ml.g6.16xlarge instance type, batch mode | $4.99 |
Dimensions summary
You pay by the hour for each running inference host (HostHrs). Pricing is organized by two choices: the GPU instance type and the inference mode. Instance types span the g5, g6, and g6e families, in sizes from xlarge up to 16xlarge. Larger sizes carry more compute and a higher hourly rate. You also pick batch mode or real-time mode. Both modes are offered for g5 and g6 instances, while g6e instances offer real-time mode only. You are not charged until an endpoint runs, and you pick the instance size that fits your workload.
Top-of-mind questions for buyers
HostHrs meters one running SageMaker inference host by the hour. You are billed for the time an endpoint stays deployed, regardless of how many requests it serves. Each active host accrues its own hourly rate. Running two hosts doubles the hourly charge.
Charges apply for each hour a host runs, not per request. An endpoint that stays deployed keeps accruing hourly charges even while idle. To stop software charges, delete or shut down the endpoint. Underlying AWS resource fees may still apply separately.
Both modes bill by running host-hour on your chosen instance type. Real-time mode keeps a persistent endpoint for live requests. Batch mode processes grouped inputs. The g5 and g6 families support both modes, while g6e instances offer real-time mode only.
isaacus.com
Helpful?
Vendor refund policy
To the maximum extent permitted by law, there are no refunds for consumption of this product.
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA) .
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
Delivery details
Amazon SageMaker model
An Amazon SageMaker model package is a pre-trained machine learning model ready to use without additional training. Use the model package to create a model on Amazon SageMaker for real-time inference or batch processing. Amazon SageMaker is a fully managed platform for building, training, and deploying machine learning models at scale.
Deploy the model on Amazon SageMaker AI using the following options:
Version release notes
Added support for g6e instances.
Additional details
Inputs
- Summary
For a user-friendly walkthrough of how to get started deploying Isaacus models on SageMaker, check out the Isaacus SageMaker quickstart guide on our docs.
This model runs on the fully air-gapped Isaacus SageMaker Model Server, which supports all the same functionality as the standard Isaacus API except that requests to the server must be proxied through the /invocations endpoint.
For example, if you wanted to send a POST request to /v1/embeddings with the data {"model": "kanon-2-embedder", "texts": ["This is a confidentiality clause."], "task": "retrieval/query"}, you would do so by sending /invocations the payload {"path": "/v1/embeddings", "method": "POST", "data": {"model": "kanon-2-embedder", "texts": ["This is a confidentiality clause."], "task": "retrieval/query"}}.
This means that minimal code changes are necessary to switch between the online Isaacus API and your own private Isaacus model deployments.
In fact, Python users can use the Isaacus SageMaker Python integration to automatically forward requests to the Isaacus API to SageMaker deployments using the standard Isaacus SDK.
Given that this is a private deployment and that authentication is managed by AWS, Isaacus API keys are not needed and are ignored.
As an embedding model, Kanon 2 Embedder currently only supports the /v1/embeddings endpoint. For more information on the arguments accepted and returned by that endpoint, please consult our API reference documentation .
- Limitations for input type
- All the same limitations applicable to the Isaacus API except for the need for an API key.
- Input MIME type
- application/json
{
"path": "/v1/embeddings",
"data": {
"model": "kanon-2-embedder",
"texts": [
"Who was the plaintiff in Mabo?"
],
"task": "retrieval/query"
}
}
{
"path": "/v1/embeddings",
"data": {
"model": "kanon-2-embedder",
"texts": [
"The plaintiff in Mabo was Eddie Mabo."
],
"task": "retrieval/document"
}
}
Input data descriptions
The following table describes supported input data fields for real-time inference and batch transform.
Field name | Description | Constraints | Required |
|---|---|---|---|
path | The path of the API endpoint being invoked (e.g., `/v1/embeddings`). | As an embedding model, Kanon 2 Embedder currently only supports the `/v1/embeddings` endpoint. | Yes |
method | The HTTP method used for the invocation (e.g., `POST`). Defaults to `POST`. | One of `POST`. | No |
headers | The HTTP headers to include in the invocation request. Defaults to `null`/`None`, in which case no additional headers are sent. | Must be a mapping of strings to strings. | No |
data | The data to be sent as the body of the invocation request. This can be any serializable object. Defaults to `null`/`None`, in which case no body is sent. | - | No |
Resources
Vendor resources
Support
Vendor support
To get in touch with our support team, you can reach out via the support form on our website: https://isaacus.com/support . We endeavor to respond within 24 hours.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
Similar products
Best-in-class legal enrichment, reranking, and embedding models, optimized for legal knowledge graphs and legal RAG.
State-of-the-art legal embedding and reranking models optimized for legal research, legal RAG, and legal document classification.
A first-of-a-kind enrichment model, capable of transforming unstructured documents into rich, highly structured knowledge graphs.
A state-of-the-art reranking model optimized for legal RAG, research, and classification, ranked first on Legal RAG Bench and the Massive Legal Embedding Benchmark (MLEB).
Customer reviews
No customer reviews yet
Be the first to review this product . We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.