What’s New With Anthropic’s Claude 3?
Anthropic revealed the next generation of Claude, a family of three different models — Opus, Sonnet, and Haiku, which are, supposedly, better than GPT4
The popular artificial intelligence start-up Anthropic has come out with the latest iteration of Claude, its famous AI chatbot. It is a family of AI models called Claude 3, unveiled on March 4. The company has claimed that the model has set new industry benchmarks across various cognitive tasks. It can even outperform prominent models like GPT 4 in some areas. The following article explores what’s new with Claude and how it fares against competitors like GPT 4 or Gemini.
What is Claude 3?
Claude is an LLM (Large Language Model) that powers the AI text generator from Anthropic. It is a company founded by a few former members of OpenAI, the name behind ChatGPT. Claude 3 is the latest offering from Anthropic. It is a family of AI models capable of advanced performance.
The chatbot can handle text, voice, and documents. It is trained on huge datasets to function in a natural, human-like conversation. These text-based conversations can handle multiple tasks like summarization, editing, decision-making, and code-writing.
This new family of models is ready to set new industry benchmarks. There are three state-of-the-art models, Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus, in ascending order, based on performance. Although Haiku is the least powerful, it is suitable for tasks that require an instant response.
Currently, Anthropic is offering Sonnet for free as it is powering the Claude.ai chatbot. However, you can avail of Claude Opus via Claude Pro services at only $20 per month. All these models come with a 200K token window.
Claude 3 Opus
The most powerful model that can handle highly complex tasks. It can deal with open-ended prompts and sight-unseen scenarios. It costs $15 per million tokens for input and $75 per million tokens for output. Its potential uses are:
- Task automation — Execute complex actions using APIs, interactive coding
- Research & Development — Research review, brainstorming, drug discovery
- Strategy — Advanced analysis, Market trends, Forecasting
Claude 3 Sonnet
The perfect balance between power and speed, Sonnet is available for free on Claude.ai. It offers strong computing at cheaper costs. The model is designed for high-end, large-scale AI systems. It costs $3 per million tokens for input and $15 for output. The use cases are:
- Data Processing — Search and retrieval of knowledge
- Sales — Product recommendation, Forecasting, Targeted Marketing
- Time-Saving — Code generation, quality control, parsing text from images
Claude 3 Haiku
The fastest model is compact and great for near-instant responses. It can answer simple questions, and handle such requests with great speed. The model costs $0.25 per million tokens for input and $1.25 for output. It can be used for the following:
- Customer Interaction
- Content Moderation
- Cost-saving
Has Claude 3 Dethroned GPT 4 and Gemini?
The company has produced great results, especially with Opus, the most powerful model. It has reportedly outperformed most competitors on common evaluation benchmarks. These evaluations include the MMLU, GPQA, and GSM8K fields. Opus has revealed near-human comprehension abilities and fluency, ahead of models like GPT 4 and Gemini.
But that’s not true only for Opus, but rather for all the models released. All three have performed exceptionally well in analysis, forecasting, content creation, code generation, and even basic conversation. The following table is a testament to the prowess of Claude 3.
What Has Improved With Claude 3?
Claude 3 has come with many improvements. Such as:
Instantaneous Results
Claude 3 is now faster than the older models of Claude 2 and 2.1, with Haiku being the fastest. It can easily read any data-dense research paper, including charts and graphs, of around 10K tokens within 3 seconds. And this speed is only going to improve much further with the official release of Haiku. Even Sonnet, the current model powering Claude.ai, is twice as fast as Claude 2. Yes, Opus has a similar speed as the older version of Claude, but in return, it is much more intelligent and advanced.
A Decreased Rate of Refusals
Older models of Claude often refused to address a prompt due to a lack of contextual understanding. However, the newer models have made significant progress in this area. Even prompts that border the model’s guardrails are less likely to get a refusal.
Enhanced Accuracy
With improvements made with prompt refusals, the accuracy of the model automatically shot up. However, this accuracy is also enhanced to maintain a highly accurate state at scale. Furthermore, Claude 3 is going to add citations, thus increasing its accuracy to a new level.
Longer Context Window and near-perfect recall memory
The best thing about Claude 3 is its context window, which is around 200K tokens. Although, the company claims it can even accept input exceeding 1 million tokens. But that feature will be reserved for select customers. Recall memory is also necessary when handling longer context. Opus has shown near-perfect recall, over 99% accurate.
How to Access Claude 3? Is It Available?
Opus and Sonnet are already available for use, via API. Developers can sign up and use these keys for their projects. Sonnet is available for free via Claude.ai chatbot and Opus via a pro subscription. Furthermore, Sonnet is also available on Amazon Bedrock and Google Cloud’s Vertex AI Model Garden.
Conclusion
Anthoripc has always been vocal regarding AI safety and responsible use. It has always refused to generate harmful or illegal content. As such, we can expect Claude 3 to be much safer. The developers have set down a set of rules that Claude 3 can’t bypass. However, it is much costlier than its peers. But for now, we can revel in the fact that Claude 3 is the most advanced AI model on the market. And it is the most significant step forward in the development of AI models and LLMs.
