Anthropic, the AI vendor second in measurement solely to OpenAI, has a strong household of generative AI fashions known as Claude. These fashions can carry out a spread of duties, from captioning photos and writing emails to fixing math and coding challenges.
With Anthropic’s mannequin ecosystem rising so shortly, 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 maintain 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 3.5 Sonnet, a midrange mannequin.
- Claude 3 Opus, Anthropic’s flagship mannequin.
Counterintuitively, Claude 3.5 Sonnet — the “midrange” mannequin — is probably the most succesful Claude mannequin for the time being. That’s virtually sure to alter as soon as the subsequent model of Opus, Claude 3.5 Opus, is launched.
All Claude fashions can analyze textual content in addition to photos, charts, graphs, and technical diagrams from paperwork. The fashions — which have a typical 200,000-token context window — also can comply with multistep directions, use instruments (e.g., inventory ticker trackers), and produce structured output in codecs like JSON.
A context window is the quantity of information 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 “implausible”). 2 hundred thousand tokens is equal to about 150,000 phrases, or a 600-page novel.
In contrast to many main generative AI fashions, Anthropic’s can’t entry the web, which means they’re not notably nice at answering present occasions questions. In addition they can’t generate photos — solely easy line diagrams.
As for the main variations between Claude fashions, Claude 3.5 Sonnet is quicker than Claude 3 Opus and higher understands nuanced and complicated directions. Haiku struggles with subtle prompts, but it surely’s the swiftest of the three fashions.
Claude mannequin pricing
The Claude fashions can be found by Anthropic’s API and managed platforms reminiscent of Amazon Bedrock and Google Cloud’s Vertex AI.
Right here’s the Anthropic API pricing:
- Claude 3.5 Haiku prices 25 cents per million enter tokens (~750,000 phrases), or $1.25 per million output tokens
- Claude 3.5 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 presents 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 by way of apps for the net, Android, and iOS, Anthropic presents 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 monthly, comes with 5x increased fee limits, precedence entry, and previews of upcoming options.
Being business-focused, Workforce — which prices $30 per consumer monthly — provides a dashboard to manage billing and consumer administration and integrations with information repos reminiscent of 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 every so often.)
Each Professional and Workforce subscribers get Initiatives, a characteristic that grounds Claude’s outputs in information bases, which may be model guides, interview transcripts, and so forth. These clients, together with free-tier customers, also can 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 corporations to add proprietary information into Claude in order that Claude can analyze the information 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 Initiatives 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 errors 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 below 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 house owners from submitting lawsuits.
Anthropic presents insurance policies to guard sure clients from courtroom battles arising from fair-use challenges. Nevertheless, they don’t resolve the moral quandary of utilizing fashions educated on information with out permission.
TechCrunch has an AI-focused e-newsletter! Enroll right here to get it in your inbox each Wednesday.
