Here is the new AI model Claude 3 from Anthropic: the goal is to surpass Gemini and ChatGPT

The sector ofartificial intelligence has been enriched with a new protagonist: we are referring to Anthropicthe company founded by some former employees of OpenAIwhich in the last few hours has announced a new family of artificial intelligence models, namely Claude 3. According to the company, Claude 3 would perform even better than the proposals of Google (Gemini) and OpenAI itself (ChatGPT), especially because it is multimodalthat is, capable of understanding texts and photos.

Specifically, according to Anthropic Claude 3 will be able to answer more questions as well as understanding longer instructions more precisely. Furthermore, this model will be able to process a greater amount of information thanks to the ability to better understand the context. The Claude 3 family, then, is made up of three models: Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus. The latter is the most “intelligent” model. Additionally, Opus and Sonnet are available on claude.ai and via its API; Haiku, however, will be launched soon.

By the way, all three models can be used for chatbots (fully automatic and data mining tasks).

Again, the new models have a lower probability of rejecting harmless requests. Previous versions of Claude, however, refused to respond to some innocuous prompts. Additionally, Anthropic highlighted how the Claude 3 models are capable of providing almost immediate results, even when analyzing complex material. In this sense, Haiku, or the smallest version of the family, is described as “the fastest and most convenient model on the market, capable of reading an article full of graphs and tables in less than three seconds”.

So, according to the company Claude 3 Opus it outperformed all rivals in benchmarking tests. In particular, the model demonstrated reasoning equal to a postgraduate level, higher than that of GPT-4 (it obtained 50.4%). Additionally, he demonstrated improved skills in solving mathematical problems, writing code, and understanding logical reasoning.

Finally, the Claude 3 models were trained on a combination of internal non-public data sets, third-party data and public data available as of August 2023. The company then specified that the training was carried out using Amazon Web Services (AWS) and Google Cloud hardware. In fact, both companies have invested in Anthropic, with Amazon paying $4 billion to the company. Claude 3 will be available on AWS’s Bedrock model library and Google’s Vertex AI.