AI startup Anthropic has released Claude 2.1, an updated version of its large language model (LLM), with a 200,000 context window. OpenAI’s standard GPT-4 only offers 8,000 context tokens, while its extended GPT-4 model has 32,000. 

A context window refers to the amount of data tokens that an LLM has been trained on. Words which are reviewed and then tagged for training by human data taggers are subsequently called tokens.

This allows the LLM to analyse large chunks of text for syntax and word order, which in turn enables LLMs to replicate human-like language responses.

Claude 2.1’s 200,000 context window is made up of over 150,000 words or 500 pages.  

Anthropic said this will help double the accuracy of its Claude model, making it easier for users to leverage the tool for longer documents. According to Anthropic, Claude 2.1 is now twice as honest, with a 30% reduction of hallucinations in its responses compared to the original Claude model. 

The chatbot’s summarisation skills have also improved in honesty, and it is 3-4 times less likely to wrongfully conclude what a document supports or claims. 

How well do you really know your competitors?

Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.

Company Profile – free sample

Thank you!

Your download email will arrive shortly

Not ready to buy yet? Download a free sample

We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below form

By GlobalData

Users are also now able to upload works to their Claude 2.1 tools, including financial statements, technical documentations like codebases and literary works.  

Whilst Claude 2.1 is now powering Anthropic’s chat interface on its website, the full 200,000 context window is only available to its subscription users. 

Anthropic was originally founded in 2021 by former OpenAI employees including the former VP of research at OpenAI, Dario Amodei. Anthropic split off to focus on building LLMs with greater size and safety. 

Whilst current LLMs are around 80% accurate in their responses, research company GlobalData forecasts in its executive briefing on AI that LLMs with 99% accuracy could be achieved in the next 10 to 30 years. 

This level of accuracy could bring the beginning of LLMs completely replacing job titles or teams, pushing AI beyond just a tool for human collaboration. 

By 2030, GlobalData forecasts the global AI market will be worth over $900bn.