Anthropic, the second-largest AI vendor after OpenAI, boasts a robust family of generative AI models known as Claude. These sophisticated models are able to perform various tasks, such as captioning images, composing emails, and solving mathematical and coding challenges.
With Anthropic continuously expanding its model ecosystem, keeping track of the capabilities of each Claude model can be challenging. To offer some clarity, we have compiled a comprehensive guide to Claude models, which will be regularly updated as new models and enhancements are introduced.
Claude models
The Claude models are named after literary masterpieces: Haiku, Sonnet, and Opus. The most recent additions to the lineup include:
- Claude 3.5 Haiku, a lightweight model.
- Claude 3.5 Sonnet, a mid-range model.
- Claude 3 Opus, Anthropic’s flagship model.
Interestingly, despite being labeled as a “mid-range” model, Claude 3.5 Sonnet currently stands out as the most capable Claude model. However, this is likely to change with the release of the next version of Opus, Claude 3.5 Opus.
All Claude models possess the ability to analyze various types of content, ranging from text to images, charts, graphs, and technical diagrams within documents. With a standard 200,000-token context window, these models can follow complex instructions, utilize tools like stock ticker trackers, and produce structured output in formats such as JSON.
The context window refers to the amount of data a model can analyze before generating new data, while tokens are smaller units of raw data. Two hundred thousand tokens roughly equate to 150,000 words, equivalent to a 600-page novel.
Unlike many other generative AI models, Anthropic’s Claude models do not have internet access, limiting their ability to provide up-to-date information on events. Additionally, they are not capable of generating images but can produce basic line diagrams.
When comparing the major differences between Claude models, Claude 3.5 Sonnet is faster than Claude 3 Opus and excels in understanding nuanced and complex instructions. While Haiku may struggle with intricate prompts, it is the quickest among the three models.
Claude model pricing
The Claude models can be accessed through Anthropic’s API and various managed platforms like Amazon Bedrock and Google Cloud’s Vertex AI.
Here is the pricing structure for the Anthropic API:
- Claude 3.5 Haiku costs 25 cents per million input tokens (~750,000 words) or $1.25 per million output tokens.
- Claude 3.5 Sonnet costs $3 per million input tokens or $15 per million output tokens.
- Claude 3 Opus costs $15 per million input tokens or $75 per million output tokens.
Anthropic provides prompt caching and batching options to achieve additional runtime savings. Prompt caching allows developers to store specific prompt contexts for reuse across API calls, while batching processes groups of low-priority model inference requests asynchronously.
Claude plans and apps
For individuals and businesses seeking to interact with the Claude models through web, Android, and iOS apps, Anthropic offers a free Claude plan with certain rate limits and usage restrictions.
Upgrading to one of Anthropic’s subscriptions removes these limitations and unlocks additional features. The current subscription options are:
– Claude Pro: Priced at $20 per month, this plan includes 5x higher rate limits, priority access, and previews of upcoming features.
– Team: Costing $30 per user per month, the Team plan is tailored for businesses and provides a dashboard for billing and user management, along with integrations with data repositories like codebases and CRM platforms such as Salesforce. It also allows toggling to enable or disable citations to verify AI-generated claims. Both Pro and Team subscribers gain access to Projects, which anchor Claude’s outputs in knowledge bases, and Artifacts, a workspace for editing and adding to content generated by Claude.
For customers with more advanced needs, Claude Enterprise offers the ability to upload proprietary data for analysis, a larger context window of 500,000 tokens, GitHub integration for syncing repositories, and access to Projects and Artifacts.
A word of caution
As with all generative AI models, there are inherent risks associated with using Claude models.
These models may occasionally make errors when summarizing information or answering questions due to a tendency to hallucinate. Trained on publicly available web data, some of which may have copyright restrictions, Anthropic and other AI vendors argue for fair-use protection, though legal challenges have arisen.
Anthropic has implemented policies to mitigate legal risks for certain customers but does not resolve the ethical dilemmas surrounding the use of models trained on potentially copyrighted data without permission.
TechCrunch offers an AI-focused newsletter! Sign up here to receive it every Wednesday.
