Anthropic has announced the availability of Claude 2, its new model which brings improved performance and longer responses. Claude 2 is available via an API which costs the same as Claude 1.3 and via the beta chat experience, which is only available in the US and UK right now.
According to the company, Claude can now receive inputs of up to 100,000 tokens which means you can give it hundreds of pages of content to analyse. It can output memos, letters, stories, and other content up to a few thousand tokens in one go too.
If you’ve not tried it, one of the defining aspects of Claude is that you can give it PDF and other text documents to look over. For example, say you’ve just been laid off by Microsoft and need to get your CV ready, you can feed your existing CV to Claude and ask for improvements and it’ll give you them.
An area where Claude 2 excels over previous versions is in coding and mathematics. In a Python coding test called Codex HumanEval, Claude 2 scored 71.2% up from 56.0% and on the GSM8K grade-school maths problems, Claude 2 scored 88.0% up from 85.2%.
Anthropic said that it has an “exciting roadmap” for capability improvements planned for Claude 2 and will be deploying them slowly and iteratively over the coming months. In terms of providing harmless responses, Claude 2 is two times better at this than Claude 1.3.
With its ability to parse documents such as PDFs, Claude 2 sets itself apart from ChatGPT, Bard, and Bing Chat. Unfortunately, you can only use the web interface in the US and UK right now but the company has promised to make its AI “more globally available” in the coming months.
If you live outside the US or UK, you may be able to get some limited access to Claude 2 via Poe.com but you will be restricted to 30 queries per day on a free account. Interestingly, you can attach files with Poe and have other AI models analyse the files too including Google"s PaLM and ChatGPT.