Anthropic’s Claude 3.5 Sonnet model is claimed to surpass GPT-4o or Gemini 1.5 Pro in many different tasks.
The Claude 3.5 Sonnet AI model has just been launched by Anthropic. Photo: Anthropic.
Anthropic, an artificial intelligence (AI) startup, has launched the Claude 3.5 Sonnet model that focuses on analyzing, creating text and images. Through benchmarking, Claude 3.5 Sonnet even surpasses OpenAI’s GPT-4o on some tasks.
According to The Verge, the Claude 3.5 Sonnet is a mid-range model. Anthropic uses the name Haiku for the small model, and Opus for the large model. However, the company claims that the Claude 3.5 Sonnet has a higher performance than the Claude 3 Opus.
Specifically, this model can analyze complex and nuanced statements, understanding concepts such as humor. The response speed of the Claude 3.5 Sonnet is twice as fast as the old version.
Visual ability is also the outstanding feature of the Claude 3.5 Sonnet when it can analyze charts and graphs more accurately, in addition to recognizing text from “unclear” images such as distortion or blur.
The amount of text that Claude 3.5 Sonnet can analyze before generating new text is about 200,000 tokens, or 150,000 words. This number is on par with the Claude 3 Sonnet.
The Claude 3.5 Sonnet’s benchmark score compares to some other models. Photo: Anthropic.
In Anthropic’s comparison table, Claude 3.5 Sonnet surpassed GPT-4o in college-level reasoning ability, writing programming code but lost in math solutions.
For vision-related tasks such as mathematical inference, scientific diagrams, or graph analysis, Anthropic’s model is more effective than GPT-4o and Gemini 1.5 Pro
Michael Gerstenhaber, Vice President of Product at Anthropic, said that the improvements to the Claude 3.5 Sonnet come from architecture tweaks and new training data, including data generated by the AI itself.
According to TechCrunch, the model training data comes from text, images, and feedback from testers. Not disclosing the details of the source of the data could help Anthropic avoid copyright litigation, which OpenAI, Google or Amazon are facing.
Artifacts feature on the Claude chatbot. Photo: Anthropic.
Besides the new language model, Anthropic also introduced Artifacts, a feature that allows users to edit and add content directly to the chatbot’s response instead of having to copy it to another application.
Anthropic said that with the new model and tools, the company wants to make Claude an app for companies to “safely bring knowledge, documents and work into a common space.”
Claude 3.5 Sonnet is released for free to users of the Cloude chatbot on the web and iOS. Subscribers who subscribe to the Claude Pro and Claude Team packages will have 5 times more interactions.