Claude

Frequently Asked Questions

Is Claude''s handling of my enterprise data secure?

Certainly, Claude follows industry-standard best practices for data handling and retention. You can find more detailed information in our Privacy Policy. Additionally, all commercial deployments are covered by Anthropic''s Data Protection Addendum, which can be provided upon request.

Where can I access the model card for Claude?

You can locate our model card for Claude 2 right here.

How does the API function?

Our API serves as a backend to seamlessly integrate Claude into any application you''ve developed. Your application sends text to our API and receives a response through server-sent events, a web-based streaming protocol. To get started, we provide API documentation with ready-to-use example code in both Python and Typescript.

What versions of Claude are currently available?

We presently offer two versions of Claude:

Claude - our most powerful model, excelling in various tasks, including sophisticated dialogue, creative content generation, and detailed instruction following.

Claude Instant - a faster and more cost-effective model that remains highly capable, suitable for tasks such as casual dialogue, text analysis, summarization, and document question-answering.

What are your pricing details?

You can access our model pricing information here.

Which languages does Claude support?

While Claude is primarily trained in English, it performs well in many other commonly used languages and can still communicate to some extent in less common languages. Additionally, Claude possesses extensive knowledge of common programming languages.

Can Claude browse the internet?

No, Claude is designed to be self-contained and doesn''t access the internet for information. However, you can provide Claude with text from the internet and request it to perform tasks with that content.

What is the purpose of constitutional training?

Constitutional training is a method used to train a model to adhere to a defined "constitution" of desired behavior. The core Anthropic model has undergone constitutional training with the goal of becoming helpful, honest, and harmless. For more information about constitutional training, you can learn more here.

What does "HHH" stand for?

"H" stands for Helpful, "H" stands for Honest, and "H" stands for Harmless (HHH). These are the three components of building AI systems like Claude that align with people''s interests. Claude aims to be:

Helpful: Claude''s primary goal is to assist the user. Honest: Claude shares information it believes to be true and avoids providing false information. Harmless: Claude does not engage in activities that could cause harm. While no existing model is perfect in terms of HHH, ongoing research aims to improve these aspects. For more information on how we evaluate HHH in our models, you can read our paper here.

How can I customize Claude''s behavior further?

You can extensively modify Claude''s behavior using prompts. Prompts allow you to explain the desired role, task, background knowledge, and provide examples of desired responses. You can find more information on prompt design here.

Can Claude models undergo fine-tuning?

In most cases, well-crafted prompts are sufficient to achieve the desired results without the time and expense of fine-tuning. However, some large enterprise users may benefit significantly from fine-tuned models. Please get in touch with us to discuss whether your needs might be best addressed with a fine-tuned model by filling out this form.

What is the context window length for Claude?

The combined context window for input and output is approximately 100,000 tokens, roughly equivalent to 70,000 words, depending on the content type.

Can Claude perform embeddings?

At this time, Claude does not perform embeddings. We have found that open-source SBERT embeddings are typically suitable for most purposes.