Anthropic launched Claude 3.5 Sonnet on Thursday, which the AI startup says outperforms previous AI models and the recently launched OpenAI model GPT-4 Omni On several scales. The company also released Artifacts, a new dynamic workspace within Claude where users can edit and build on Claude’s AI projects, like a playable crab video game.
“Our goal with Claude is not to create an incrementally better MBA, but rather to develop an AI system that can work alongside people and software in meaningful ways,” Dario Amodei, co-founder and CEO of Anthropic, said in a press release. “Features such as artifacts are early experiments in this direction.”
This is the first release in Anthropic’s Claude 3.5 model family, coming just three months Launch of Cloud 3 models. Anthropic is struggling mightily to keep up with OpenAI’s product launches. Today, Anthropic is just launching version 3.5 of its mid-tier Sonnet model, but the startup plans to release version 3.5 of the Haiku (its entry-level model) and Opus (the more capable version) later this year. The startup also says it is exploring features like web search and memory for future releases.
Anthropic’s focus is not just on releasing better models, but on releasing more features and capabilities to make its AI more useful. Using Artifacts, Anthropic says users can create interactive projects within Claude. In the demo, the startup demonstrates how you can create characters and icons within an AI-generated 8-bit crab video game, editing the project along the way.
The AI startup says Claude 3.5 Sonnet is its most powerful vision model to date, which allows Claude to perform better at visual reasoning, interpreting diagrams, and transcribing text from imperfect images. Anthropic claims that the Claude 3.5 Sonnet outperforms many of the vision capabilities of the GPT-4 Omni, which debuted impressive vision capabilities at launch. Claude is supposedly better at understanding diagrams, documents, and mathematics visually. Otherwise, Anthropic says Claude’s new software generally outperforms OpenAI’s ChatGPT in programming and reasoning.
Notably, Anthropic says that the Claude 3.5 Sonnet does a better job of analyzing what to answer and what not to answer. AI chatbots have become notorious for refusing to answer certain questions. Gizmodo did a deep dive into AI censorship a few months ago, and found that Anthropic’s Claude refused to answer a lot of questions. With CLOUD 3.5, Anthropic feels confident in CLOUD’s judgment to answer inappropriate or potentially harmful questions.
