Anthropic on Friday released the first member of the Claude 3.5 large language model (LLM) family, Claude 3.5 Sonnet. The new release comes just three months after the AI company unveiled the Claude 3 family of artificial intelligence (AI) models. According to the company, the latest AI model can outperform Claude 3 Opus, the most capable model of the previous generation. It also gains a new Artifact feature that allows users to see a sandbox view when the chatbot is asked to generate code snippets, website designs and more.
Claude 3.5 Sonnet Details
According to a news editorial after by the company, the Claude 3.5 Sonnet is now more powerful and cost-effective than the Claude 3 Opus. According to Anthropic’s naming convention, the Haiku model is the smallest, the Sonnet model is in the middle, and Opus is the flagship-level AI model. The AI company is also expected to release the 3.5 Haiku and Opus models in the coming months.
As for the performance of the new AI model, the Claude 3.5 Sonnet comes with a context window of 200,000 tokens and costs $3 (approximately Rs. 250) per million input tokens and $15 (approximately Rs. 1,254) per million output tokens. The company claims that the latest model can run twice as fast as the Claude 3 Opus model. It scored 88.7 percent in 5-shot (AI was trained on a small number of labeled examples) on the Massive Multitask Language Understanding (MMLU) benchmark and 92.0 percent in 0-shot on the HumanEval benchmark.
Anthropic claimed that the Claude 3.5 Sonnet can write, edit, and execute code autonomously. It also has reasoning and problem-solving capabilities that allow it to perform complex tasks such as code translation.
In addition to the text- and coding-based upgrades, the new AI model also gets improved computer vision. The AI company claimed that the Claude 3.5 Sonnet can transcribe text from imperfect images, graphs or illustrations. A new Artifact feature is also included in the new AI chatbot. It’s a sandbox-style preview window that opens when the AI is asked to generate content, such as code snippets, text documents or website designs. Users can view, edit and build upon Claude’s generation.
In terms of safety parameters, Anthropic stated that Claude 3.5 Sonnet is still at the ASL-2 (AI Safety Level 2) standard. According to the AI company’s Responsible Scaling Policy, ASL-2 refers to systems that show early signs of dangerous capabilities (such as issuing instructions to create a bioweapon), but are not considered useful due to a lack of reliability and that the information is already publicly available.
In addition to the security rating, Anthropic also claimed that it brought in external experts to test and improve security mechanisms with Claude 3.5 Sonnet. The AI model was also provided to the UK’s Artificial Intelligence Safety Institute (UK AISI) for evaluation before it was deployed.
Currently, Claude 3.5 Sonnet is available for free on Claude.ai and the Claude iOS app. Claude Pro and Claude Team subscribers also have access with higher speed limits. It is also available via Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI.
