If you’ve been keeping up with AI news, you’ve heard of ChatGPT and Claude. These two AI chatbots have been constantly at the cutting edge and creating news stories since they came out. They’ve also been proven particularly useful for professionals in many industries.
ChatGPT and Claude both claim to offer the best AI interactions that benefit humanity. However, they offer different features and benefits that set them apart. In this comparison, we will:
Explain what ChatGPT and Claude are
Highlight the main features that set them apart
Compare their costs and capabilities
What Is ChatGPT?
ChatGPT is an AI chatbot released to the public by OpenAI in November 2022. It was the first of the now mega-popular language models used as chatbots.
Depending on what you want from it, ChatGPT is a diverse multi-purpose tool. Its first significant uses were answering questions, explaining ideas, generating new ideas, and helping with problem-solving tasks. Over time, ChatGPT models have gotten even better at problem-solving coding issues or giving text-to-speech explanations for hundreds of topics.
ChatGPT is used for both entertainment and professional purposes. It can become a complete digital assistant, trained on the information you provide it. Or, it can just be a high-IQ virtual friend with whom to make jokes or have discussions.
All of this happens from a deceptively simple interface. It looks little different from a social media chat feature, but behind that interface, ChatGPT is an artificial neural network designed to understand and communicate with human languages. Yes, it’s also a great translator.
Specifically, ChatGPT was built on top of a transformer model that processes sequenced data. Basically, it pays attention to different parts of text input while breaking it down. But it doesn’t function quite like, or as well as, a human brain.
How ChatGPT Works
ChatGPT is one of the most significant Large Language Models (LLMs), now trained on over 1.8 trillion parameters. Parameters are weights that AI systems are trained to apply to data. With LLMs like ChatGPT, these parameters help it understand the truly complex aspects of language, including nuance, context, patterns, and peculiarities like sarcasm.
The vast “knowledge” that ChatGPT seems to demonstrate is the result of extensive, deliberate training and fine-tuning. Each GPT model was trained on a massive mix of books, websites, and additional texts. Then, there was the first stage of unsupervised learning, followed by supervised learning.
The “unsupervised learning” process saw ChatGPT left to itself and forced to sort through unlabeled data. So, it read the text data to find structure and patterns, then decoded it without specific instructions. This was the part where ChatGPT became able to simulate brilliance, soaking up data in a way no person could. That included all of the nuances of human language present in the materials it was trained on.
Next, ChatGPT was fine-tuned to use all of its understanding to perform specific tasks. These include the controls and guardrails meant to keep interactions relevant and safe. Combining these two approaches leads to complete interactions that end users experience.
With all of this training, ChatGPT’s transformer architecture can understand our languages at a level that seems similar to our understanding. It’s what makes the final outputs ChatGPT generates match the context of your inputs so well. It’s what creates the “human” feeling that ChatGPT simulates.
Transformer architecture involves an encoder and a decoder:
Encoder: Processes an input sequence to build a contextually coherent understanding using all of the positions in the input.
Decoder: Generates output based on the encoder’s contextual understanding and previously generated tokens. Outputs are produced one at a time.
As with autoregressive language models, ChatGPT uses just the second part of the transformer architecture to understand not only context but also nuance and tone.
In essence, ChatGPT takes input text and breaks it down into smaller units, including words and parts of words (a process called “tokenization”). Those smaller units are the tokens that the decoder uses to produce outputs one token at a time.
What Is Claude?
Claude is an AI chatbot released by Anthropic in March 2023. It’s one of the most popular and highest-performing AI chatbots, leading some people to wonder what the difference between Claude and ChatGPT is.
If you skipped the above sections about ChatGPT, we suggest you read them to understand how LLMs like ChatGPT and Claude work. The underlying technology is essentially the same, but there are significant differences in Claude. These are:
The total number of parameters
The training materials
The safeguards against certain kinds of content
But like ChatGPT, Claude is also based on transformer architecture and can understand:
Words
Relationships between words
The context of whole sentences
Language patterns and nuance
This same process powers both of these generative AI chatbots as well as others like Gemini and Grok.
Like ChatGPT, Claude was trained on a massive quantity of text information available online. For both models, the training information comes from a wide range of sources.
In addition to the enormous process of training an AI model, Claude offers some unique features that differentiate it from ChatGPT—and all of the other chatbots.
Constitutional AI
Image credit: Anthropic
Anthropic’s mission statement focuses on ensuring AI ends up benefiting individuals and society. Much of the language focuses on harm reduction. This mission is addressed with a second model based on “Constitutional AI” (CAI).
In short, CAI is artificial intelligence with solid guardrails. In other words, Anthropic baked in fine-tuning to reduce any potential harm that could arise from using their tool.
Typically, AI researchers tackle harm reduction through an ongoing human review process to eliminate the chances of negative outcomes. Parameters are controlled to not allow illegal or unethical content wherever possible. This is done from the early stages of training and continues after the chatbot has gone public. The reinforcement learning from human feedback (RLHF) model usually is applied to enable users to add more restrictions.
Anthropic’s Claude models use CAI in a way that’s similar to RLHF to apply guardrails geared toward high-stakes use cases. For example, AI models that carry higher risks are less appropriate as legal or healthcare assistants. Thus to reduce risk, Claude rejects directions to steer outputs into an inflammatory direction.
Although Claude’s CAI applies something similar to the RLHF model, Claude focuses more on building a self-enforcing set of principles cemented by the platform’s users. According to Anthropic’s blog, weekly updates involving their analysis and human feedback take place.
So, the main issue here is what constitutes “inflammatory” content. Does Claude censor anything that a user complains about? Or does it let a lot of potentially offensive content slip through?
According to Anthropic, the CAI model promotes “normative” principles written into the company’s constitution. User feedback and employee reviews provide a lot of input. However, the CAI framework is based on training the model on strong moral authorities outside the company. That means training the Claude models on the Universal Declaration of Human Rights, for example.
Beyond training the model to view human rights law as a guide, Anthropic also works with a platform called Polis to develop a collective constitution based on the feedback of a massive audience representing all demographics.
So, Claude’s models have a unique approach to powerful and transparent guardrails. They may not be perfect models that fit with all users’ needs; however, CAI is a strong, constantly developing framework that differentiates Claude from other AI models.
The Main Differences Between ChatGPT and Claude
ChatGPT and Claude have similar underlying technology, but they differ in their missions. Granted, they both identify part of their missions as leveraging AI to benefit humanity. However, they differ in terms of the process of steering AI in a beneficial direction.
Despite this seemingly large difference, ChatGPT and Claude have a lot in common. In some ways, they are more alike than either one is with other AI chatbots.
Cost
Both Claude and ChatGPT offer extensive free services to users. All you need is an email address to sign up and start using either (or both) at no cost.
If you want to get the best possible quality out of your chatbot, Claude Pro and ChatGPT Pro both cost USD $20. It’s beyond this point that prices may differ. Enterprise plans for either Claude or ChatGPT can be customized. Using various premium features extensively implies a much wider cost range.
Company
OpenAI is, relatively speaking, a veteran in the AI industry. Having been founded in 2015, it consistently led the way and maintained the most users with its ultra-popular GPT models.
Anthropic is a bit younger, being launched in 2021. The company was founded by a team of former OpenAI employees, including a former OpenAI Vice President of Research. The company has continued to stay up-to-date and highly competitive with Claude models that add unique benefits to AI chatbot use.
Both companies are among the best established in the industry and have secured billions of dollars in funding from other AI giants.
Context window
An AI chatbot’s context window is its textual range around a target token. Think of it as the processing and generation speed. OpenAI’s Help resources state that the GPT-4 models have context windows of up to 128,000. One news article published on Anthropic’s website states Claude 3 models have context windows of up to 200,000.
What this means is that Claude 3 has a better memory of the tokens (pieces of information) from a conversation. This is significant for users who want to:
Have longer conversations
Get the chatbot to perform tasks with larger bodies of text (think medical books, studies)
Use the chatbot for enterprise-level tasks
Have a chatbot generate much longer responses
Images
Claude models are currently unable to generate images but ChatGPT offers an image generator through its regular interface. You can get ChatGPT to create any appropriate-for-work image with a series of text prompts. The more numerous and detailed your prompts are, the better the image will be.
Languages
ChatGPT understands and communicates in over 50 languages. It fluently speaks all major world languages and some regional dialects with decent proficiency. Its multilingual abilities are largely because of the massive quantity of data and the many languages used to train ChatGPT.
Claude can also speak many languages but is highly proficient in only a dozen. According to Anthropic, Claude is exceptionally well-trained in Portuguese, French, and German. Other languages are still understood, but Claude may not have the specialized training to understand complex topics.
In both cases, the focus of LLM training was overwhelmingly on the English Language.
ChatGPT vs. Claude on Accuracy
ChatGPT and Claude have offered some of the highest levels of accuracy. Each of their most recent models is among the leaders in overall knowledge, math, and coding.
GPT-4o | Claude 3.5 Sonnet | |
MMLU | 88.7% | 88.7% |
HumanEval | 90.2% | 92% |
ARC * | 96.4% | 91% |
Credit: paperswithcode.com
* GPT-4 vs. Claude 2
As for MMLU, the broadest accepted knowledge benchmark, GPT-4o and Claude 3.5 Sonnet got caught in a tie. Their older and free versions also performed very well against the other top chatbots.
MMLU tests AI systems with multiple-choice questions in 57 subjects, including STEM, the humanities, and social sciences. Claude and ChatGPT have a lot of knowledge and their outputs are highly accurate, performing well against the relevant human experts.
HumanEval tests each model’s code generation abilities. Claude’s top-performing model beats ChatGPT’s by a small margin, but both are leaders in that field as well.
The ARC benchmark tests common-sense problem-solving. It tests general knowledge to a large extent, but more importantly, it tests chatbots for their understanding of nuance and their ability to deduce solutions to problems. In this case, data for older versions of ChatGPT and Claude are available, and both are still among the top performers, although GPT-4 proved a bit closer to perfect at common-sense problem-solving than Claude 2.
Overall, if you’re comparing accuracy between ChatGPT and Claude, you’re starting in the right place! Both models have consistently produced accurate responses in a wide range of subjects relative to other companies’ models.
None of this is to say that either should be considered an authority in any particular subject. Both companies include disclaimers in all their models’ interfaces. It’s your responsibility to fact-check any information that a chatbot produces for you. Any inaccurate or plagiarized information you submit is on you, regardless of what part AI played in it. The same goes for poor code or anything else.
Similarities Between ChatGPT & Claude
ChatGPT and Claude take different approaches to controls and are trained on various parameters. However, while those differences can produce very different outcomes, the two chatbots have many similarities.
As the end user, you need to keep a few things in mind regardless of which of these two chatbots you use.
Originality
AI chatbots, including ChatGPT and Claude, may produce content that’s unoriginal. That means they may produce text that’s repetitive, empty, or plagiarized. The extent to which AI-generated content is problematic will vary based on:
The model that you’re interacting with
Your previous prompts
The task you assign to it
The reason why Claude, ChatGPT, and others may give you plagiarized information stems from their training. To make each of them the extremely knowledgeable tools they are, their parent companies trained them on a lot of data. The training that enables AI chat tools to provide such impressive responses includes a mix of:
Material in the public domain
Readily available online information and “common knowledge”
Copyright-protected content
This diversity is required to make AI chatbots dynamic and able to tackle such a wide range of user queries. But it can also create two problems: originality and accuracy issues.
Most AI chatbots have been through lawsuits over allegations that training material was used without permission, compensation, or copyright infringement over uses of protected sources, such as:
Academic research
News articles
Published works of fiction and nonfiction
Well-known works of art
In one example, at the end of 2023, eight newspapers sued OpenAI and Microsoft over the alleged use of news articles in training. The allegations included “millions” of articles used without payment or permission.
This means that it can be risky for any AI chatbot user to submit AI-generated work to schools or employers haphazardly. The end user is responsible for such submissions. Misusing that information can constitute copyright infringement, plagiarism, and other problems.
Anyone can now use AI detectors, like the appropriately named AI Detector, to quickly find out how likely it is that a text was written by AI.
Using an AI detector will help you identify problematic places in your writing so you can revise those sections in your own words to make them more original and not appear to have been generated by AI.
Accuracy
While Claude and ChatGPT are among the most accurate AI models, the accuracy remains less than perfect. If you want to use ChatGPT or Claude, make sure you fact-check any claims they generate against reputable sources. Using ChatGPT or Claude to help with ideas is fine, but make sure to research each significant point’s factual accuracy.
ChatGPT vs. Claude: The Last Word
Between ChatGPT and Claude, there isn’t a clear winner. Going forward, these two trailblazers are likely to differentiate themselves even more while competing to be the best AI chatbots. Claude just launched a refreshed Claude 3.5 Haiku and Sonnet, while ChatGPT just launched a new Windows app. That’s to say nothing about Orion, which may come out any time during the next few months.
It’s hard to keep up with the breaking news and the rapid improvements Anthropic and OpenAI have made. That said, both have consistently kept their chatbots at the top of performance benchmarks and remain constantly relevant.
ChatGPT may continue to dominate in terms of performance in several areas, but when compared with other chatbots, Claude is perhaps in the best position to compete. Claude will also continue to stand out for its Constitutional AI, which may win it more users in the long run.
In the end, both AI chatbots can provide responsible users with lightning-speed assistance for contextual, mathematical, and programming tasks. Using ChatGPT or Claude responsibly leads to similar benefits. Ultimately, they both offer some of the best free models for you to try, and their Pro plans also cost the same amount. So why not try both for free and upgrade to the pro plan for the one that works best for you?
