March 2025 Summaries

The Anthropic Economic Index has released its second research report, covering data from Claude 3.7 Sonnet, the company's newest and most capable model with strengths in agentic coding and an "extended thinking" mode. The results show a rise in usage for coding, educational, scientific, and healthcare applications, as well as increased use of the extended thinking mode predominantly in technical and creative problem-solving contexts. The report also analyzes augmentation vs. automation patterns across tasks and occupations, finding that automation is still dominating in some categories, while others are highly augmentative. Additionally, a new bottom-up taxonomy of usage on Claude.ai has been released, covering 630 granular categories ranging from household plumbing to digital art creation. These datasets are freely available for researchers to download.

Claude can now be used to search the internet and access more up-to-date information, providing direct citations for fact-checking. This enhancement expands Claude's knowledge base with real-time insights, enabling it to provide answers based on current data. Users can leverage web search in various ways, including sales planning, financial analysis, research, and purchasing decisions. The feature is currently available as a preview for paid users in the United States, with plans to expand support to free users and additional countries soon.

AI models are rapidly advancing across various domains, including cybersecurity and biology, with Claude, a frontier AI model, demonstrating significant improvements in cyber capabilities, but still lagging behind human experts. While the models have made substantial progress, they fall short of thresholds at which they generate substantially elevated risks to national security. The Frontier Red Team's work provides valuable insights into the trajectory of potential national security risks from frontier AI models and highlights the importance of internal safeguards, independent evaluation entities, and targeted external oversight to ensure responsible development of AI. The team is committed to scaling up their evaluations and risk mitigations, moving with urgency to address the rapidly advancing capabilities of AI models, and exploring deeper collaboration between frontier AI labs and governments to improve their work.

Anthropic has responded to the draft report of the California Governor's Working Group on AI Frontier Models, welcoming their emphasis on transparency as a means to create a well-functioning AI policy environment. The company agrees that transparency can be a low-cost, high-impact way to grow the evidence base around new technologies and increase consumer trust. Anthropic adheres to industry best practices in its own responsible scaling policy, which includes publicly disclosing model testing results and performing third-party testing. However, the company believes that governments could play a constructive role in improving transparency in the safety and security practices of frontier AI companies, particularly by requiring them to have a safety and security policy and publicly document their tests. Anthropic looks forward to providing further feedback on the report and helping shape California's approach to frontier model safety.

The Anthropic API has made several updates to optimize throughput and reduce token usage with the release of Claude 3.7 Sonnet, including cache-aware rate limits, simpler prompt caching, and token-efficient tool use. These updates aim to help developers process more requests within their existing rate limits while reducing costs. The new features include prompt caching, which reduces costs by up to 90% and latency by up to 85%, as well as a text_editor tool designed for collaborative document editing workflows. Early users such as Cognition are already leveraging these updates to improve token efficiency and response quality. These features are available today with minimal code changes, allowing developers to take advantage of the new capabilities.

Anthropic has submitted a comprehensive set of recommendations to the Office of Science and Technology Policy (OSTP) in response to the White House's request for information on an AI Action Plan. The recommendations aim to better prepare America to capture the economic benefits and national security implications of powerful AI systems, which are expected to emerge around 2026 or 2027. These systems will possess intellectual capabilities rivaling those of Nobel Prize winners and will be able to navigate all digital interfaces, control technology instruments, reason through complex tasks, and interface with the physical world. Anthropic's submission focuses on six key areas: national security testing, strengthening export controls, enhancing lab security, scaling energy infrastructure, accelerating government AI adoption, and preparing for economic impacts. The recommendations aim to strike a balance between enabling innovation and mitigating risks posed by increasingly capable AI systems.

The Anthropic Console has been redesigned to serve as a single platform for building, testing, and iterating on AI deployments with Claude and teammates. Developers can now share prompts directly within the console, collaborate more efficiently, and control the extended thinking budget. The console also includes tools to write, evaluate, and optimize prompts, as well as automatically generate and refine them. Additionally, it provides a centralized way for teams to develop, refine, and standardize prompts across their organization, making it easier to establish best practices and ensure consistent quality. The console now supports the latest Claude 3.7 Sonnet model and allows users to control the extended thinking budget, optimizing prompts for this new capability.

Anthropic has raised $3.5 billion at a post-money valuation of $61.5 billion, led by Lightspeed Venture Partners and other investors, to advance its AI system development, expand compute capacity, and accelerate international expansion. The company plans to further develop its coding abilities with the launch of Claude 3.7 Sonnet and Claude Code, aiming to create AI systems that collaborate with teams to tackle complex projects and achieve significant impact. Businesses across industries are leveraging Anthropic's solutions, including fast-growing startups and global corporations, to transform their operations and drive revenue growth. With this investment, Anthropic remains committed to deepening its understanding of frontier AI systems and ensuring that artificial intelligence advances human progress.