You’ve probably heard of both Bard and ChatGPT by now. However, another highly capable chatbot burst onto the scene earlier this year called Claude. It comes courtesy of Google and Amazon-backed startup Anthropic, which was only founded in 2021.
Claude is definitely one of the most impressive chatbots to be launched in the last 12 months. More than 350,000 people signed up to the waitlist to use Claude 2, the most recent version of the language model.
In this guide, you'll learn exactly what Claude is, how it works, and all about its parent company, Anthropic. You'll also see how Claude differs from ChatGPT in key areas.
- What is Anthropic AI?
- Who Owns Anthropic, and Who Funds It?
- What is Claude AI? An Intro to the Chatbot
- 8 Questions with Claude [Test]
- Claude Pricing & Tokens Explained
- Claude 2 and ChatGPT: Key Differences
- Should You Use Claude AI?
What is Anthropic AI?
Anthropic is an AI startup company based in San Francisco, California. The company describes itself as an AI “safety and research” business and focuses focused on creating “reliable, interpretable, and steerable AI systems.”
It was founded in 2021 by Dario and Daniela Amodei, and now has more than 150 employees.
Anthropic is classed as a public-benefit company – a corporation that is set up to make a profit, but is doing so by working on something that will have a positive impact on humanity.
🔎 Want to browse the web privately? 🌎 Or appear as if you're in another country?
Get a huge 86% off Surfshark with this special tech.co offer.
Anthropic largely focuses its work on building large language models (LLM) and the chatbots that subsequently use them. Its most famous offering so far is Claude 2, which underpins the sophisticated chatbot of the same name.
Who Owns Anthropic, and Who Funds it?
Anthropic is still headed up by founders Dario and Daniela Amodei, who are CEO and President respectively. They were previously senior figures at OpenAI, who owns ChatGPT.
GPT-3 engineer Tom Brown and fellow Co-founder Sam McCandlish are also among the ranks, along with other former OpenAI employees.
During Anthropic's first funding round, which took place in early 2021, the company raised an impressive $124 million – which included investment from Skype Co-founder Jaan Tallin.
Since then, Google has invested almost $400 million in the startup and holds a 10% stake in the company, while Amazon has pledged up to $4 billion. As part of the latter deal, Amazon employees and cloud customers will get early access to the technology.
What is Claude AI? An Intro to the Chatbot
Claude is an AI assistant (i.e. chatbot) developed by Anthropic. It was first released in March 2023, powered by the language model Claude 1.3. However, a second version – powered by a language model called Claude 2 – was released in July 2023.
Anthropic also says Claude 2 scores higher on Bar tests and that it was twice as good at giving “harmless” responses. However, its dataset was cut off in 2022 and it can't connect to the internet, so it struggles to give accurate answers relating to events that happened after that point in time.
Anthropic has also released another, nimbler, lighter model that gives quicker responses, called Claude Instant. The company says that this version of Claude is lighter and quite a bit faster than Claude 2, but that Claude 2 is better at complex reasoning and the more powerful offering overall.
Claude expanded its “context window” from 9,000 tokens to 100,000 tokens in May 2023 shortly before releasing Claude 2. This means it can handle inputs of around 75,000 words, so businesses can submit huge reams of documentation for Claude to analyze.
Should You Use Claude AI?
Having written about the AI revolution and the expansion of the chatbot ecosystem for some time now, I've used Claude regularly since its release.
Aside from Bard and ChatGPT, I'd say it’s probably the most capable chatbot I've tested. Its answers are more cogent than the likes of Character AI, and it’s actually got a really sleek, welcoming interface.
If you’d like to find out how Claude 2, Bard and various chatbots compare, check out our guide to the best ChatGPT alternatives.
8 Questions to Ask Claude 2
To give you a little bit of a taste of what Claude is like to use and what the interface looks like, I asked the chatbot a series of different questions. Here's how it responded:
Claude Pricing & Tokens Explained
You can use Claude 2 for free, but you’ll be limited. Now, you can purchase Claude Pro – which was first released in September 2023 – for $20 per month. This makes it the same price as ChatGPT Plus, the paid-for version of the popular chatbot that will let you install ChatGPT plugins, among other paywalled features.
Claude Pro will give you five times as much usage as the free plan, as well as early access to new Claude features.
However, if you’re going to be using Claude a lot, there’s another way you can pay for access to the standard version of Claude as well as the “Claude Instant” version discussed above.
The cost of Claude and Claude Instant, in this case, will depend on how many tokens you buy. “Tokens” are very tiny elements of a language model – they can correspond to words, subwords, characters, and bytes, says Anthropic, which describes them as “atoms”. When you input a prompt into Claude, it's then converted into tokens.
Claude Instant tokens cost $1.63 per million for prompts and $5.51 per million for completion (i.e. outputs). Claude 2 tokens cost $11.02 per million for prompts and $32.68 per million for completion. Here’s the information on Claude pricing that Anthropic provides:
Claude AI and ChatGPT: Key Differences
Claude is a robust and well-engineered AI chatbot, and it's a good option to have in your back pocket if ChatGPT goes down. but there are some key differences between Claude 2 and ChatGPT that are good to know about before you start using either chatbot.
Language model
Claude is the name given to both the Chatbot and the large language model developed by Anthropic. It is trained on over 137 billion text and code parameters – the same as Meta's Llama 2.
Parameters are different variables that are learned by an LLM during training, ones that can be adjusted to sculpt the way it replies to responses. Generally, more parameters mean a more powerful language model.
ChatGPT’s free version uses GPT-3.5, trained on 175 billion parameters – so it’s not that much larger than Claude 2. GPT-4, which powers the paid version, is trained on an enormous set of 1.5 trillion parameters.
Data retention and usage
Since ChatGPT burst onto the scene in 2022, privacy advocates have been demanding more information about how the chatbot utilizes user data. OpenAI has always been open about the fact that ChatGPT saves data, that user conversations might be used for AI training purposes, and that these may even be seen by humans working at the company. However, the company will retain your conversations for up to 30 days after you delete them.
Anthropic's Claude will save your conversations too. The company says that it will “automatically delete prompts and outputs on the backend within 90 days of receipt or generation unless you and we agree otherwise”, and that it won't “use your conversations in our consumer or beta/evaluation services to train our model.” The only exception is if you're flagged for a trust and safety review.
Response moderation
One big difference between ChatGPT and Claude 2 is that Claude is generally considered better at consistently producing safe responses. The reason? Claude 2 is built with something called Constitutional AI.
Claude is given a set of values that its answers have to adhere to and can then fine-tune itself using the constitution, rather than waiting for human feedback and input, which is how ChatGPT learns what responses it should refuse to answer.
In one phase of model training, Claude was asked to “critique and revise its own responses using the set of principles and a few examples of the process”. In the second phase, reinforcement learning is used to train the model further, but as we've covered, it generates its own feedback for itself using the constitution.
Context window size
As we mentioned above, Claude 2 has a context window of 100K tokens, or around 75,000 words – which means you can, in theory, upload large reports, legal documents, etc.
GPT-3.5 Turbo has a context window of 4,097 tokens, while GPT-3.5-16K has a context window of about 16,385 tokens. GPT-4, which is only available to ChatGPT Plus customers, has a context window of more than 32,000.
Subscribers to rival AI chatbot Perplexity’s Pro account can now also use Claude’s long context window of 100,000.
Revenue
A final big difference is the company's revenues. Anthropic predicts to be making around $17 million in monthly revenue by the end of 2023 and hopes to generate around $500 million annually by the end of 2024.
ChatGPT, on the other hand, is creeping towards $1 billion annual revenue. It’s used by more people and now has a premium version itself. However, this doesn’t necessarily mean either company is hyper-profitable. It costs around $700,000 per day just to keep ChatGPT up and running and responding to user queries.
