Anthropic, one of many world’s largest AI distributors, has a strong household of generative AI fashions known as Claude. These fashions can carry out a spread of duties, from captioning photographs and writing emails to fixing math and coding challenges.
With Anthropic’s mannequin ecosystem rising so rapidly, it may be powerful to maintain observe of which Claude fashions do what. To assist, we’ve put collectively a information to Claude, which we’ll hold up to date as new fashions and upgrades arrive.
Claude fashions
Claude fashions are named after literary artworks: Haiku, Sonnet, and Opus. The newest are:
- Claude 3.5 Haiku, a light-weight mannequin.
- Claude 3.7 Sonnet, a midrange, hybrid reasoning mannequin. That is presently Anthropic’s flagship AI mannequin.
- Claude 3 Opus, a big mannequin.
Counterintuitively, Claude 3 Opus — the most important and most costly mannequin Anthropic affords — is the least succesful Claude mannequin in the meanwhile. Nonetheless, that’s certain to alter when Anthropic releases an up to date model of Opus.
Most not too long ago, Anthropic launched Claude 3.7 Sonnet, its most superior mannequin so far. This AI mannequin is totally different from Claude 3.5 Haiku and Claude 3 Opus as a result of it’s a hybrid AI reasoning mannequin, which may give each real-time solutions and extra thought of, “thought-out” solutions to questions.
When utilizing Claude 3.7 Sonnet, customers can select whether or not to activate the AI mannequin’s reasoning talents, which immediate the mannequin to “assume” for a brief or lengthy time period.
When reasoning is turned on, Claude 3.7 Sonnet will spend anyplace from just a few seconds to some minutes in a “pondering” part earlier than answering. Throughout this part, the AI mannequin is breaking down the consumer’s immediate into smaller elements and checking its solutions.
Claude 3.7 Sonnet is Anthropic’s first AI mannequin that may “motive,” a way many AI labs have turned to as traditional methods of improving AI performance taper off.
Even with its reasoning disabled, Claude 3.7 Sonnet stays one of many tech trade’s top-performing AI fashions.
In November, Anthropic launched an improved – and costlier – model of its light-weight AI mannequin, Claude 3.5 Haiku. This mannequin outperforms Anthropic’s Claude 3 Opus on a number of benchmarks, however it may possibly’t analyze photographs like Claude 3 Opus or Claude 3.7 Sonnet can.
All Claude fashions — which have a typical 200,000-token context window — may also comply with multistep directions, use tools (e.g., inventory ticker trackers), and produce structured output in codecs like JSON.
A context window is the quantity of knowledge a mannequin like Claude can analyze earlier than producing new information, whereas tokens are subdivided bits of uncooked information (just like the syllables “fan,” “tas,” and “tic” within the phrase “improbable”). 200 thousand tokens is equal to about 150,000 phrases, or a 600-page novel.
Not like many main generative AI fashions, Anthropic’s can’t entry the web, that means they’re not significantly nice at answering present occasions questions. Additionally they can’t generate photographs — solely easy line diagrams.
As for the key variations between Claude fashions, Claude 3.7 Sonnet is quicker than Claude 3 Opus and higher understands nuanced and complicated directions. Haiku struggles with subtle prompts, nevertheless it’s the swiftest of the three fashions.
Claude mannequin pricing
The Claude fashions can be found via Anthropic’s API and managed platforms corresponding to Amazon Bedrock and Google Cloud’s Vertex AI.
Right here’s the Anthropic API pricing:
- Claude 3.5 Haiku prices 80 cents per million enter tokens (~750,000 phrases), or $4 per million output tokens
- Claude 3.7 Sonnet prices $3 per million enter tokens, or $15 per million output tokens
- Claude 3 Opus prices $15 per million enter tokens, or $75 per million output tokens
Anthropic affords immediate caching and batching to yield further runtime financial savings.
Immediate caching lets builders retailer particular “immediate contexts” that may be reused throughout API calls to a mannequin, whereas batching processes asynchronous teams of low-priority (and subsequently cheaper) mannequin inference requests.
Claude plans and apps
For particular person customers and corporations trying to merely work together with the Claude fashions through apps for the net, Android, and iOS, Anthropic affords a free Claude plan with fee limits and different utilization restrictions.
Upgrading to one of many firm’s subscriptions removes these limits and unlocks new performance. The present plans are:
Claude Professional, which prices $20 per thirty days, comes with 5x greater fee limits, precedence entry, and previews of upcoming options.
Being business-focused, Staff — which prices $30 per consumer per thirty days — provides a dashboard to regulate billing and consumer administration and integrations with information repos corresponding to codebases and buyer relationship administration platforms (e.g., Salesforce). A toggle allows or disables citations to confirm AI-generated claims. (Like all fashions, Claude hallucinates sometimes.)
Each Professional and Staff subscribers get Tasks, a function that grounds Claude’s outputs in data bases, which could be type guides, interview transcripts, and so forth. These clients, together with free-tier customers, may also faucet into Artifacts, a workspace the place customers can edit and add to content material like code, apps, web site designs, and different docs generated by Claude.
For purchasers who want much more, there’s Claude Enterprise, which permits firms to add proprietary information into Claude in order that Claude can analyze the data and reply questions on it. Claude Enterprise additionally comes with a bigger context window (500,000 tokens), GitHub integration for engineering groups to sync their GitHub repositories with Claude, and Tasks and Artifacts.
A phrase of warning
As is the case with all generative AI fashions, there are dangers related to utilizing Claude.
The fashions often make mistakes when summarizing or answering questions due to their tendency to hallucinate. They’re additionally educated on public net information, a few of which can be copyrighted or beneath a restrictive license. Anthropic and lots of different AI distributors argue that the fair-use doctrine shields them from copyright claims. However that hasn’t stopped information homeowners from filing lawsuits.
Anthropic offers policies to guard sure clients from courtroom battles arising from fair-use challenges. Nonetheless, they don’t resolve the moral quandary of utilizing fashions educated on information with out permission.
This text was initially revealed on October 19, 2024. It was up to date on February 25, 2025 to incorporate new particulars about Claude 3.7 Sonnet and Claude 3.5 Haiku.