Claude Instant outperforms leading AI chatbots with 100,000 tokens
Anthropic has released a new version of Claude that accepts 100,000 tokens, or around 75,000 input words.
This allows users to analyze and perform tasks on large excerpts from books, codes, documents, transcripts and more.
Screenshot of Poe, June 2023Read on to learn more about the latest version of the Anthropic AI chatbot, Claude-instant-100k, and see how it performs tasks compared to its competitors: ChatGPT, Bing AI, and Google Bard.
Restrictions on AI chatbot input
Do you want to ask questions about content? Here’s the length of input you can send to generative AI chatbots from OpenAI, Microsoft, and Google.
- OpenAI’s standard GPT-4 model offers 8,000 context tokens, while the extended context length model offers 32,000 tokens / 25,000 input words.
- The Bing AI Chat prompt input field shows 0/4000, indicating that only up to 4,000 characters are accepted. However, the chatbot accepted 40,400 characters from Lorem Ipsum and accurately counted it as a total of 6,000 words.
- Google Bard accepts up to 10,000 characters.
Claude-instant-100k accepts – three times more input than the best GPT-4 model.
But how well can they follow directions?
I experimented with Claude-instant-100k, ChatGPT Plus, Bing AI and Google Bard with three tasks. Here are the results.
Task #1: Edit a 1,580-word list using Claude, ChatGPT, Bing, and Bard
The input: A list of 126 items (some of the first ChatGPT plugins) at 1,850 words / 12,500 characters.
The Prompt: Can you rephrase and edit the descriptions for the following list:
[the list]
Claude completed the task
- Claude finished editing my list at item #126 without hallucinations and followed the instructions to rephrase the item descriptions.
- Compared to the other AI chatbots, Claude-instant-100k completed the task the fastest.
ChatGPT Default GPT-4 Hallucinating (link to chat)
- ChatGPT stops in the middle of point #37.
- I clicked continue generating – ChatGPT finished #37 and stopped again.
- I asked it to continue please.
- ChatGPT starts listing items that weren’t in my original input.
- After a few more clicks to continue generating, ChatGPT finished editing my list at point #142.
Bing AI tried
- Bing accepted the entire prompt and list.
- Was asked if this was what I was looking for – I said yes and please continue.
- After another prompt to please continue, Bing stopped editing my list at item #126 with no hallucinations.
- Unfortunately, after point #18, the descriptions weren’t rephrased anymore and the list was outputting as I was typing it.
Bard was unable to create images
- Due to the 10,000 character input limit, I was only able to submit 99 items from the list in the prompt.
- Bard replied, “I can’t create images yet, so I can’t help you with that.” or “That’s something I can’t do yet.”
Ultimately, Claude-instant-100k completed the task with the greatest accuracy and speed.
You can compare the output of Claude, ChatGPT and Bing with the original list of items in it Google sheet.
Task #2: Summarize a 23,500-word PDF using Claude, ChatGPT, Bing, and Bard
The input: 23,508 words / 125,104 characters, copied from a PDF documenting TikTok’s responses to a hearing on how Congress can protect US citizens from the social platform.
The prompt: Please summarize the following:
[the text]
Claude summarized the entire text succinctly
Screenshot of Poe, June 2023ChatGPT-Standard-GPT-4 requested something shorter
Screenshot of ChatGPT, June 2023Bing AI lost connection
Screenshot from Bing, June 2023Bard summarized the first 10,000 characters
Screenshot from Google Bard, June 2023Task #3: Edit this PHP code using Claude, ChatGPT, Bing and Bard
The input: A PHP file with 14,787 characters.
The Command Prompt: Can you debug the following:
[the code]
Claude made some suggestions
Screenshot of Poe, June 2023ChatGPT GPT-4 code interpreter found it ok
Screenshot of ChatGPT, June 2023Bing AI could not debug the code
Screenshot from Bing, June 2023Bard considered the first 10,000 characters error-free
Screenshot from Google Bard, June 2023Claude-instant-100k outperforms the competition
In just a few initial experiments with everyday tasks, it becomes clear that Anthropic’s latest version of Claude can outperform its competitors.
As with ChatGPT, the experimental features for Claude come at a price.
Poe subscribers can use Claude instant 100k in a desktop browser for $19.99 per month, comparable to ChatGPT+ subscriber prices.
Screenshot of Poe, June 2023If you have tasks that require lengthy typing, try the new version of Claude.
Featured image: Koshiro K/Shutterstock
