Anthropic stands as a leading force in the AI field, boasting an impressive suite of generative AI models known collectively as Claude. These models can undertake various tasks, ranging from generating image captions and composing emails to tackling mathematical problems and coding issues.
As the ecosystem of Anthropic’s models expands rapidly, keeping up with the specific capabilities of each Claude model can be quite challenging. To assist, we’ve created a comprehensive guide to Claude, which we will continue to update as new models and enhancements are developed.
Overview of Claude Models
The Claude models draw their names from notable pieces of literature, including Haiku, Sonnet, and Opus. Currently available models are:
- Claude 3.5 Haiku, a compact model.
- Claude 3.7 Sonnet, a versatile, mid-tier reasoning model currently serving as Anthropic’s flagship offering.
- Claude 3 Opus, a robust model.
Interestingly, although Claude 3 Opus is the largest and highest-priced model from Anthropic, it currently holds the least capabilities among the Claude lineup. This is expected to change with forthcoming updates to Opus.
Recently, Anthropic introduced Claude 3.7 Sonnet, its most sophisticated model yet. Distinct from Claude 3.5 Haiku and Claude 3 Opus, Sonnet is a hybrid reasoning model capable of delivering both immediate and more contemplative responses to inquiries.
Users of Claude 3.7 Sonnet can decide whether to engage the model’s reasoning abilities, which allow it to ponder for either a brief or extended period before providing a response.
When reasoning is activated, Claude 3.7 Sonnet may take several seconds to a few minutes in a “thinking” phase prior to responding. During this period, the AI dissects the user’s request and verifies its answers.
Claude 3.7 Sonnet represents Anthropic’s inaugural model that can reason, a method increasingly adopted by AI research labs as traditional enhancement strategies lose effectiveness.
Even without reasoning engaged, Claude 3.7 Sonnet remains among the top-tier AI models in the tech sector.
In November, Anthropic unveiled an enhanced and pricier version of Claude 3.5 Haiku. This model surpasses Claude 3 Opus in various benchmarks, though it lacks the image analysis capabilities of Claude 3 Opus and Claude 3.7 Sonnet.
All Claude models feature a standard context window of 200,000 tokens, enabling them to follow complex, multi-step instructions, utilize tools (like stock ticker trackers), and produce outputs in structured formats such as JSON.
The context window refers to the amount of information a model like Claude can process before generating new content, while tokens represent individual segments of raw data (like the syllables “fan,” “tas,” and “tic” in “fantastic”). A total of 200,000 tokens is approximately equivalent to 150,000 words or the content of a 600-page novel.
Unlike many leading generative AI models, Anthropic’s offerings are unable to access the internet, which limits their proficiency in addressing questions related to current events. Moreover, they can only produce simple line illustrations rather than full images.
When comparing the Claude models, Claude 3.7 Sonnet operates more quickly than Claude 3 Opus and exhibits a superior understanding of intricate and nuanced prompts. Conversely, while Haiku is the fastest of the three, it struggles with more complex tasks.
Pricing for Claude Models
The Claude models will soon be available via Anthropic’s API along with managed platforms such as Amazon Bedrock and Google Cloud’s Vertex AI.
Here’s the pricing structure for the Anthropic API:
- Claude 3.5 Haiku is priced at $0.80 per million input tokens (~750,000 words) and $4 per million output tokens.
- Claude 3.7 Sonnet is available for $3 per million input tokens and $15 per million output tokens.
- Claude 3 Opus costs $15 per million input tokens and $75 per million output tokens.
Anthropic provides prompt caching and batching options for additional runtime efficiency.
Prompt caching enables developers to save specific “prompt contexts” to reuse in subsequent API calls, while batching processes groups of low-priority model requests together, making them more cost-effective.
Plans and Applications for Claude
For individuals and organizations looking to interact with Claude models through web, Android, and iOS applications, Anthropic offers a complimentary Claude plan with usage limits and restrictions.
Upgrading to one of Anthropic’s subscription plans removes these limitations and introduces new features. Available options include:
- Claude Pro.
- Claude Team, tailored for small businesses.
- Claude Enterprise, designed for larger organizations.
Claude Pro, priced at $20 per month, comes with five times the standard rate limits, priority access, and previews of forthcoming features.
Targeted at businesses, the Team plan is available for $30 per user monthly, including a dashboard for billing management and user control, plus integrations with data repositories such as code repositories and customer relationship management systems (e.g., Salesforce). This plan features a toggle to enable or disable citations for validating AI-generated outputs (as all models can occasionally produce inaccurate information).
Both Pro and Team subscribers can access Projects, a feature that enables Claude’s outputs to be connected to knowledge bases, such as style guides and interview transcripts. Users at any subscription level also have access to Artifacts, a workspace designed for editing and enhancing content like applications, designs, and documents produced by Claude.
For those requiring even more capabilities, Claude Enterprise allows companies to upload proprietary data for analysis. It includes a larger context window of 500,000 tokens, GitHub integration for teams to synchronize their repositories, in addition to Projects and Artifacts.
Cautionary Notes
As with all generative AI models, using Claude comes with certain risks.
These models can sometimes make errors in summarization or answering questions due to their propensity to hallucinate. Additionally, they are trained on publicly available web data that may sometimes be under copyright or restricted licenses. Although Anthropic and many other AI companies assert that the fair use doctrine protects them from copyright disputes, this has not deterred data owners from filing lawsuits against them.
Anthropic does offer policies to shield select customers from potential legal challenges arising from fair-use disputes. However, these do not address the ethical issues tied to the use of models trained on data without the consent of its authors.
This article was initially published on October 19, 2024, and was updated on February 25, 2025, to incorporate the latest information on Claude 3.7 Sonnet and Claude 3.5 Haiku.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
