Anthropic Launches Claude 3.5 Sonnet: Setting New AI Benchmarks and Introducing 'Artifacts'

Anthropic has officially released Claude 3.5 Sonnet, setting a new benchmark for intelligence in the rapidly evolving AI landscape and introducing innovative features for user interaction. Positioned as the first release in their forthcoming Claude 3.5 model family, this new Sonnet model surprisingly outperforms the company's previous top-tier model, Claude 3 Opus, on a wide array of evaluations while operating at twice the speed.

A New Benchmark in Intelligence and Speed

Claude 3.5 Sonnet demonstrates significant improvements across key industry benchmarks. Anthropic reports substantial gains in graduate-level reasoning (GPQA), undergraduate-level knowledge (MMLU), and coding proficiency (HumanEval). The model showcases a more nuanced understanding of context, humor, and complex instructions, along with improved quality and tone in its writing.

Notably, these intelligence gains come with a major performance boost – Claude 3.5 Sonnet operates at twice the speed of Claude 3 Opus, making it ideal for complex, context-sensitive tasks like customer support and multi-step workflow orchestration.

Introducing Artifacts: A New Way to Interact

Alongside the model, Anthropic unveiled Artifacts, a new feature on Claude.ai that enhances user productivity. When Claude generates content like code snippets, text documents, or website designs (e.g., using SVG), these elements now appear in a dedicated "Artifacts" window next to the chat. This creates a dynamic workspace where users can see, edit, and build upon Claude's creations in real-time, integrating AI assistance more seamlessly into their workflows. Anthropic views this as a step towards collaborative work environments.

State-of-the-Art Vision Capabilities

The new model also boasts Anthropic's most advanced AI vision capabilities to date, surpassing Claude 3 Opus on standard vision benchmarks. It shows remarkable improvement in interpreting charts and graphs, and accurately transcribing text even from imperfect or distorted images – a core capability for industries relying on visual data analysis like retail, logistics, and finance.

Availability and Pricing

Claude 3.5 Sonnet is available immediately through Anthropic's API, on the web at Claude.ai, and via the Claude iOS app. It is free to use on Claude.ai, with subscribers to Claude Pro and Team plans benefiting from significantly higher rate limits.

Crucially, despite its superior performance, Claude 3.5 Sonnet is priced affordably – matching the cost of the previous generation Claude 3 Sonnet ($3 per million input tokens, $15 per million output tokens, with a 200K token context window). This makes it significantly more cost-effective than Claude 3 Opus for most use cases. It's also becoming available on platforms like Google Cloud and Amazon Bedrock soon.

Safety and What's Next

Anthropic emphasizes that Claude 3.5 Sonnet maintains high safety standards, having undergone rigorous testing and evaluation by internal teams and external bodies like the UK's Artificial Intelligence Safety Institute (AISI). Safety improvements have reportedly made it better at refusing inappropriate requests compared to previous models.

This release marks the beginning of the Claude 3.5 generation, with Claude 3.5 Haiku and Claude 3.5 Opus set to launch later this year, aiming to complete the family and further push the balance of intelligence, speed, and cost.

Claude 3.5 Sonnet represents a significant leap forward, offering state-of-the-art intelligence and practical new features at an accessible price point, setting a new standard for generally available AI models.

Read the full announcement on the Anthropic Blog.