Anthropic, flush with funds from Amazon and Google, has made its Claude chatbot service available in an iOS app and introduced a business-oriented Teams plan.

"Early testers report that the Claude app is exceptional for brainstorming ideas on the go, getting quick answers to questions, or analyzing scenes and images from the real world," the startup announced. "Our new iOS app puts the power of frontier intelligence in your back pocket."

Now "frontier intelligence" sounds like something an AI model would write. In this context it means "industry-leading machine learning," rather than deep knowledge of the American West. The Claude 3 family of large multimodal models, which debuted in March, "outperforms its peers on most of the common evaluation benchmarks for AI systems," according to Anthropic.

At least that was the case until April when Meta introduced its open source LLaMA 3 model, which surpasses comparable Claude models in some benchmarks.

In any event, the iOS app is free to use - up to a point. The app includes a button to upsell users on a Pro subscription for $20 per month, which gets you "at least 5x the usage compared to our free service," according to the support documentation.

The amount of text that entails isn't precisely defined, but Anthropic suggests that Pro tier users will be able to submit 45 messages consisting of about 200 English sentences (15-20 words each) every five hours before limitations kick in.

A cursory traffic analysis of the Claude app using Proxyman for iOS suggests Anthropic imposes a default hard limit of 25,000 tokens for its free tier and 190,000 for its Pro tier. The free tier also looks like it has a 4096 token output cap. A token is typically about four English characters.

An Anthropic spokesperson didn't immediately respond to a request to confirm those details.

Pro users can choose from any of the three Claude models, Opus (most capable), Sonnet or Haiku (fastest). Free tier defaults to the Sonnet model, a spokesperson told us.

What might one do with Claude on iOS? Well, you could resume a web chat with Claude that's been synced to your mobile device. You might also ask the chatbot to summarize a document, or analyze a picture, if you need help with that sort of thing.

Your results may vary. Presented with a photo of a Doberman Pinscher sleeping on a dog bed, Claude offered a hedged, not entirely accurate description: "The dog appears to be a Doberman Pinscher or similar breed, with a sleek black coat and pointed ears."

In fact, the dog's ears were not pointed because they were not cropped and hung down over the edge of the bed. Presumably Claude went with "pointed" because that would be statistically common for that breed.

But mostly accurate is par for the course when playing with frontier intelligence. As the app itself warns, "Claude can make mistakes. Please double check responses."

Claude's Teams plan is slightly more expensive at $30 per month. That gets even more usage than the Pro plan, access to the full Claude 3 model family, a context window that can handle 200,000 tokens (long documents), administrative and billing management tools, and higher priority access when Anthropic's servers are under load.

Anthropic also plans to add more collaboration features shortly, "including citations from reliable sources to verify AI-generated claims, integrations with data repositories like codebases or CRMs, and iterating with colleagues on AI-generated documents or projects." ®

https://www.theregister.com//2024/05/01/anthropic_claude_ios_teams/