Anthropic recently launched its new family of generative AI models under the umbrella term “Claude 3.” The family essentially includes three distinct variants of the LLM in a bid to compete with larger competitors such as Gemini and ChatGPT. Tech behemoths like Amazon and Google have generously funded Anthropic—the startup behind Claude. Claude 3 is supposedly more intelligent, capable, and safer when compared to its predecessors, such as Claude 2.1, Claude 2, and its base version. Anthropic’s Claude series, which is known for its prioritization of safety, has also gone multimodal in this new release, with its models now capable of processing both text and image prompts.
Given that the AI market has gotten immensely competitive in the recent past with the release of capable models such as the GPT-4 Turbo and Google Gemini, firms like Anthropic, which have also maintained a reputation for quality, are looking to get ahead with their new launches. Claude 3 is touted to be the most intelligent of the firm’s LLMs and is bound to revolutionize generative AI possibilities for users of large language models. Claimed to be better than both GPT-4 and Gemini LLMs, Anthropic aims to outperform its competitors on various fronts. The upcoming sections of the article take a closer look at Claude 3 and its variants.
Claude 3’s Key Features: Anthropic’s First Multimodal AI Model
Claude 3 has impressive multimodal capabilities that allow for data analysis and more.
Claude 3 has been launched with three variants: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. The last variant is by far the most advanced of them all and has numerous capabilities. Since the Claude 3 models are multimodal, they can process both image and text prompts. Interestingly, the latest large language model series from Anthropic can also accept up to 20 images in a single prompt, making it the only LLM to be capable of the same. Competitor models such as GPT-4 and Google’s Gemini accept only a single image per prompt. Moreover, Claude 3 models can also process slideshows, PDFs, and images in various formats. While Claude’s latest generative AI model is multimodal, Anthropic has not enabled the image generation feature, unlike Dall-E 3 and Imagen-2, which are integrated with ChatGPT and Gemini, respectively. Additionally, the model also does not recognize individuals in photographs as an ethical measure.
Claude 3 is more stable and coherent when compared to its predecessors, and it performs better than GPT 4 and Gemini on various benchmarks. The model is more expressive and generates high-quality responses, which are an improvement over older editions, making the LLM a potent option for use cases that require precise AI-generated content. However, despite its many leaps forward, Anthropic admits that even its most advanced model is prone to the occasional instance of AI hallucination. Regardless, the key tweaks in the new model enable users to generate better responses even if their engineered prompts are shorter and more succinct as compared to those deployed in legacy versions.
Claude 3’s Technical Features
Anthropic’s focus on responsible AI makes Claude a relatively safe set of LLMs.
Claude 3 is a versatile and capable model that has outperformed GPT-4 and Google Gemini Ultra 1 models on numerous key benchmarks, including the famous Undergraduate Level Knowledge test (MMLU) and the Grade School Math test (GSM8k). The new LLM model also has some of the largest context windows across the spectrum, allowing for 200,000 tokens, which is equivalent to 150,000 words at the start. Select customers of the model are also slated to get up to 1 million token-context windows, or 700,000 words. This indicates an enhanced ability to perform analytics on relevant prompts that might require the LLM to mine information or in extensive AI writing use cases. Though there isn’t much information on the dataset that Claude 3 uses, it is understood that the information contained in the dataset has a cutoff date of August 2023. Since Claude 3 isn’t connected to the internet like its predecessors, users can expect accurate information on events up to the cutoff date.
Claude 3 is also multilingual in addition to its proficiency in English. However, the model isn’t as fluent in other languages since the firm admits that it has limited data resources for other languages and dialects. Regardless, Anthropic’s Claude 3 will be regularly updated in the coming months to ensure the model becomes even more adept at handling newer challenges in the AI world. Presently, the Claude 3 Sonnet and Opus variants are available on the LLM’s web platform and can also be accessed through APIs, in addition to Google’s Vertex AI platform and Amazon Bedrock.
Claude 3 and the Prospects for Advanced Generative AI Models
Anthropic has launched the Claude Sonnet and Opus variants, while the Haiku model awaits release.
Claude presents a major jump for Anthropic in the industry, given that the new family of models outdoes some of its major competitors. Anthropic has also come out with a pricing model for its new models, which is also rather competitive. The Opus model costs $15 per million input tokens while being priced at $75 per million output tokens. Other variants, such as Sonnet, are priced at $3 per million tokens of input and $15 for the same number of tokens of output. The base Haiku mode, on the other hand, will be priced lower at $0.25 and $1.25 per million tokens of input and output, respectively. Since Claude follows the novel principles of “constitutional AI,” the firm emphasizes responsible AI heavily. This is especially important, given that regulations such as the European Union’s AI Act will soon be in place. That being said, safety is just as critical, and firms like Anthropic producing highly capable models like Claude 3, show promise for the ethical AI industry.
FAQs
1. When was Claude 3 launched?
Claude 3 was launched on March 4, 2024. The model is a successor to Anthropic’s other LLMs such as Claude 2 and 2.1.
2. Is Claude 3 free?
No, Claude 3 is not free and will be priced based on the firm’s new pricing model for the latest LLM. The Opus model costs $15 per million input tokens while being priced at $75 per million output tokens. Other variants, such as Sonnet, are priced at $3 per million tokens of input and $15 for the same number of tokens of output. The base Haiku mode, on the other hand, will be priced lower at $0.25 and $1.25 per million tokens of input and output, respectively.
3. How many variants does Claude 3 have?
Claude 3 has three variants—Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. The last model is the most advanced of them all.