OpenAI’s ChatGPT now has a new rival. In the past few hours Anthropicthe artificial intelligence company founded in 2021 by seven former OpenAI employees, has in fact announced Claude 3.5 Sonnethis new large language model or LLM (Large Language Model). According to what was declared by the San Francisco startup, in some internal tests the new model surpassed the results obtained by models from competing companies – including OpenAI’s GPT-4o, Google’s Gemini 1.5 Pro and Meta’s Llama 3 400B – as well as that the results obtained by its high-end model, i.e Claude 3 Opus.
Claude 3.5 Sonnet is already available to Claude.ai users on the web and on the Claude app for iOS/iPadOS. The company also announced that the model is also available throughAnthropic API, Amazon Bedrock And Vertex AI by Google Cloud. The model is expensive $3 per million input tokens And $15 per million output tokenswith a contextwindow of 200,000 tokens.
How Claude 3.5 Sonnet works and what it can do: the tests passed
The tests passed by Claude 3.5 Sonnet have highlighted several positive aspects, at least according to what Anthropic declared. According to the company, Claude 3.5 Sonnet shows a significant improvement in capturing nuances, humor and even complex instructions, and is quite agile in high-quality content writing using a tone that Anthropic called “natural and relatable.” Since Claude 3.5 Sonnet acts a double the speed of Claude 3 Opusit is perfect for use in complex activities, such ascustomer service.
In putting its new model to the test, Anthropic said it put the model through a test that said it would fix a bug or add functionality to an open source code base. Claude 3.5 Sonnet managed to solve 64% of the problemsagainst the 38% success achieved by Claude 3 Opus.
The company also stated that the model is able to interpret the information present in the images quite accurately, reporting the following:
The Claude 3.5 Sonnet is our most powerful vision model yet, surpassing the Claude 3 Opus on standard vision benchmarks. These breakthrough improvements are most noticeable for tasks that require visual reasoning, such as interpreting charts and graphs. Claude 3.5 Sonnet can also accurately transcribe text from imperfect images, a key feature for retail, logistics and financial services, where AI can glean more information from an image, graph or illustration than from the text only.
At the same time as the announcement of Claude 3.5 Sonnet, Anthropic also presented Artifacts (available on Claude.ai), which is a new feature that expands the way users can interact with the model. In practice, when you query the latter asking for example to generate content such as text documents, website projects, code fragments, etc., the outputs are shown in a dedicated window next to the chat with the model. This way, the user can interact with a dynamic workspace. According to Anthropic, “this preview feature marks Claude’s evolution from a conversational AI to a collaborative work environment.”
When it could be released: Antropic’s next moves
Anthropic also announced that it is already working to complete the Claude 3.5 family of models – Claude 3.5 Haiku And Claude 3.5 Opus – which will be released by 2024. Beyond that, the company is “developing new ways and features to support more use cases for businesses, including integrations with enterprise applications.”
