Unlocking Creativity: Exploring Claude 3.7 Sonnet and Claude Code for Enhanced Content Generation

We’re excited to introduce Claude 3.7 Sonnet¹. This new model is smarter than ever, combining fast responses with in-depth reasoning. Users can choose to see step-by-step thought processes, and API users can set limits on how long Claude can take to think.

Claude 3.7 Sonnet excels in coding and web development tasks. We’re also launching a new tool called Claude Code that helps developers tackle complex coding jobs directly from their command line. It’s currently in a limited research phase.

You can access Claude 3.7 Sonnet across all Claude plans, including Free and Pro, as well as through the Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI. The extended thinking feature is available on most plans, though not the free tier.

Pricing remains the same as previous models: $3 per million input tokens and $15 per million output tokens, which includes thinking tokens.

Claude 3.7 Sonnet: Smart Reasoning Made Practical

We designed Claude 3.7 Sonnet to think both quickly and deeply, just like humans. This unifies its response capabilities, creating a smoother experience for everyone. In standard mode, it builds on Claude 3.5 Sonnet. When switched to extended thinking mode, it engages in more thoughtful processing, which enhances its ability to tackle math, science, coding, and other complex tasks. Users find that prompting works similarly in both modes.

With the API, you can also set a thinking budget, allowing Claude to focus on quality or speed based on your needs.

Our team has shifted our focus to real-world applications, making sure Claude is even better for everyday tasks instead of just theoretical problems.

Early tests show Claude leading the pack in coding skills. From handling complex codebases to advanced tool use, it’s proven to be unmatched. Our collaborations with companies like Cursor and Vercel show that Claude can create production-ready code and tackle full-stack updates effectively.

Introducing Claude Code

Since June 2024, developers have preferred Sonnet for their projects. Today, we’re enhancing this experience with Claude Code, our new coding tool currently in a limited preview.

Claude Code assists developers with tasks like searching through code, editing files, running tests, and pushing updates to GitHub, while keeping users informed along the way.

Though still in its early stages, Claude Code is already crucial for our team, particularly in debugging and testing. In early tests, it has completed work in short bursts that would normally require over 45 minutes of manual effort.

We will keep improving Claude Code based on usage, focusing on reliability and tool support.

Our aim is to learn how developers interact with Claude to improve future models. By participating in this preview, you’ll gain access to tools we use and help shape its future.

Collaborating with Claude on Your Code

We’ve upgraded the coding experience on Claude.ai. Our GitHub integration is available on all Claude plans, allowing developers to connect their repositories directly to Claude.

Claude 3.7 Sonnet offers a powerful way to improve your projects. It understands your work better than ever, making it easier to fix bugs and create documentation.

Responsible Development

We’ve rigorously tested Claude 3.7 Sonnet to ensure it meets high standards for security and reliability. This model is better at distinguishing harmful requests, leading to a 45% reduction in unnecessary refusals compared to earlier versions.

Our system card details the new safety evaluations and how Claude is trained to handle emerging risks, particularly prompt injection attacks. It also discusses the advantages of reasoning models in understanding decision-making and ensuring reliability. Full details are available in the system card.

Future Prospects

Claude 3.7 Sonnet and Claude Code represent significant advances in AI. Their ability to think deeply and work alongside humans brings us closer to a future where AI significantly enhances human potential.

We’re eager for you to explore these new tools. Your feedback is valuable as we refine and enhance our models.