Anthropic, a company backed by Google and Amazon, yesterday launched a new generation of its artificial intelligence model: Claude 3. This family of models sets new standards in various cognitive tasks. With Claude 3, Anthropic hopes to compete with OpenAI’s ChatGPT-4.
Here’s an overview of the key features of Claude 3
The Three Claude 3 Models
The Claude 3 Models consist of three distinct versions: Haiku, Sonnet, and Opus. It aims to address businesses’ biggest generative AI concerns: cost and performance.
Claude 3 Haiku
The company claims it is the fastest and most cost-effective model on the market for its intelligence category. It can read a dense research article with graphics and tables in less than three seconds.
Claude 3 Sonnet
Twice as fast as Claude 2 and Claude 2.1, this model can handle tasks that require quick responses, such as knowledge retrieval or sales automation.
Claude 3 Opus
According to Anthropic, this model outperforms its peers on most common AI system evaluations. It achieves levels of understanding and fluency close to that of a human, paving the way for advanced general intelligence.
In a technical paper, Anthropic claims that Claude 3’s three models beat OpenAI’s GPT-3.5 and Gemini 1.0 Pro in knowledge, reasoning, math, problem-solving, and coding. They can also converse in different languages, English but also Japanese or French.
Here is a comparison of the 3 models to those of the peers on various benchmarks of capability:
Advanced Vision Capabilities
Claude 3 models possess sophisticated vision capabilities, comparable to other leading models. They can process a variety of visual formats such as photos, graphs, technical diagrams, etc.
This feature is particularly valuable for businesses whose knowledge bases contain up to 50% of information encoded in various formats, such as PDFs, organizational charts, or presentation slides.
We tested it and the result is quite impressive:
The company also insists on the fact the responses the models bring are more trustworthy than previous models.
In a press release, Anthropic said that
“The Claude 3 family of models will initially offer a 200K context window upon launch. However, all three models are capable of accepting inputs exceeding 1 million tokens and we may make this available to select customers who need enhanced processing power.”
Applications and Price
Opus
The cost of utilizing Opus ranges from $15 to $75 per million tokens, depending on the specific input and output requirements.
With a context window of 200K, the potential uses of this technology are diverse. It can be employed for task automation, facilitating the planning and execution of complex actions across various APIs and databases, as well as interactive coding.
It also has applications in research and development, enabling tasks such as research review, brainstorming, hypothesis generation, and even drug discovery.
Besides, it proves beneficial in strategic endeavors, allowing for advanced analysis of charts, graphs, financials, market trends, and forecasting.
Sonnet
Claude 3 Sonnet is well-suited for enterprise applications. With a cost ranging from $3 to $15 per million tokens and a context window of 200K, it provides a cost-effective solution without compromising on performance.
Potential uses include data processing tasks such as RAG or search and retrieval across extensive knowledge bases.
The model also proves valuable in sales-related activities like product recommendations, forecasting, and targeted marketing.
Claude 3 Sonnet is also valuable for time-saving tasks such as code generation, quality control, and parsing text from images.
Haiku
Claude 3 Haiku is designed to deliver near-instantaneous responsiveness. It is priced between $0.25 to $1.25 per million tokens and is equipped with a context window of 200K.
This model excels in various applications, including customer interactions such as rapid and accurate support in live chats and translations.
Moreover, it proves valuable in content moderation tasks, efficiently identifying risky behavior or customer requests.
Additionally, Claude 3 Haiku contributes to cost-saving initiatives like optimized logistics, inventory management, and extracting knowledge from unstructured data.
Competing with OpenAI
Anthropic is another company aiming to compete with OpenAI. In France, MistralAI recently released Mistral Large, a model that they say is nearly as effective as GPT-4.
