Anthropic, the AI startup backed by tech giants Google and Amazon.com, has unveiled its latest artificial intelligence model, Claude 3.5 Sonnet. This release comes just three months after the debut of its Claude 3 series.
The new model, Claude 3.5 Sonnet, is designed to surpass its predecessor, Claude 3 Opus, which was lauded by CEO Dario Amodei as the “Rolls-Royce of models.”
Claude 3.5 Sonnet boasts significant improvements, scoring higher on benchmark exams, operating at twice the speed, and being offered to developers at a fifth of the cost.
This makes it an attractive option for software developers looking for high-performance AI solutions at a competitive price.
“AI models are a bit more fungible than cars,” said Amodei quoted by Reuters. “I don’t have to buy them and hold onto them for 20 years. That’s one advantage of our field.” This flexibility allows for rapid advancements and continuous improvements, a trend mirrored by other industry leaders like OpenAI and Google.
For consumers, Anthropic has made its cutting-edge technology accessible for free via Claude.ai and an iOS app. Users can also opt into a new feature called “Artifacts,” which organizes the content generated by Claude—be it a novel outline or a simple computer game—into a user-friendly window display alongside their chat with the AI.
This feature, coupled with a new group subscription plan, is a step towards fostering collaborative work environments.
“Artifacts is about being able to work collaboratively and using your model to produce finished products,” Amodei noted.
Anthropic plans to continue this rapid development pace, with future releases including Claude 3.5 Haiku and Claude 3.5 Opus expected later this year.
“We want to have as fast a release cycle as we can, again, subject to our safety values,” said Amodei.
Safety and reliability are paramount, with the company committed to ensuring that its systems are not only capable but also aligned with human values.
The Claude 3.5 Sonnet has demonstrated impressive performance in internal evaluations. It solved 64% of problems in an agentic coding evaluation, a significant improvement over the 38% solved by Claude 3 Opus.
It also excels in nuanced communication, humor, and complex instruction capabilities, delivering high-quality content in a natural, relatable tone.
In graduate-level reasoning, Claude 3.5 Sonnet scored 59%, outperforming ChatGPT-4o’s 53%, and it achieved an 87% score in reasoning over text, surpassing ChatGPT-4o’s 83%, Google’s Gemini at 74%, and Meta’s Llama large language model, also at 83%.
Despite these advancements, Claude 3.5 Sonnet faces stiff competition.
On math problem solving, it was outperformed by ChatGPT-4o, which was 5% more accurate. The race among generative AI companies is intense, with Anthropic, OpenAI, Google, and others releasing new models at breakneck speed.
This rapid rollout has raised concerns about the potential for misuse and the introduction of biases in AI models before developers can address them.
“Creating systems that are not only capable but also reliable, safe, and aligned with human values is a complex challenge,” Amodei acknowledged. “We don’t have all the answers, but we’re dedicated to working on these problems thoughtfully and responsibly.”
In addition to new model releases, Anthropic introduced the Artifacts feature on Claude.ai, creating a dynamic workspace where users can see, edit, and build upon their projects in real-time. This marks a significant evolution for Claude from a conversational AI to a collaborative work environment.
Anthropic envisions Claude.ai expanding to support team collaboration, enabling organizations to securely centralize knowledge, documents, and ongoing work in one shared space, with Claude acting as an “on-demand teammate.”
Anthropic was founded in 2021 by siblings Dario and Daniela Amodei, both former OpenAI executives, driven by a commitment to AI safety.
The AI landscape continues to evolve rapidly, and with Claude 3.5 Sonnet, Anthropic aims to remain at the forefront, delivering innovative solutions that are as powerful as they are responsible.
