Anthropic Publishes Second Economic Index Report on Claude 3.7 Sonnet Usage
Anthropic released a second Economic Index report analyzing how people use Claude 3.7 Sonnet, including its new extended thinking mode and task-level automation data.
Quick answer
What did Anthropic's second Economic Index report say about how people are using Claude 3.7 Sonnet?
Anthropic's second Economic Index report found rising Claude.ai usage in coding, education, science and healthcare after the launch of Claude 3.7 Sonnet. Extended thinking mode was used most by technical occupations. Augmentation held steady at 57% of usage, and Anthropic released a new 630-category usage taxonomy plus task-level automation data.
Key takeaways
- Anthropic says the share of Claude.ai usage in coding, education, science and healthcare rose in the 11 days after Claude 3.7 Sonnet launched, compared with its first Economic Index report.
- According to Anthropic, extended thinking mode was used most in technical tasks, with roughly 10% usage among computer and information research scientist tasks and about 8% among software developer tasks.
- Anthropic reports that augmentation still made up 57% of Claude.ai usage, unchanged from the prior report, while the share of learning-type interactions rose from about 23% to about 28%.
- Anthropic found that copywriter and editor tasks showed the highest task iteration, while translator and interpreter tasks showed among the highest directive usage, where the model completes work with minimal human involvement.
- Anthropic released a new bottom-up taxonomy of 630 usage categories built with its Clio analysis tool, alongside task- and occupation-level automation/augmentation datasets, all available for download.
Anthropic has published a second report from its Anthropic Economic Index, an initiative the company says regularly releases data and research on how AI affects labor markets and the economy. This installment covers Claude.ai usage data collected after the launch of Claude 3.7 Sonnet, which Anthropic describes as its newest and most capable model, with strengths in agentic coding and a new "extended thinking" mode.
What Anthropic analyzed
Anthropic says it examined 1 million anonymized Claude.ai Free and Pro conversations from the 11 days following the launch of Claude 3.7 Sonnet. The company used its privacy-preserving analysis tool, Clio, to map each conversation to one of roughly 17,000 tasks in the U.S. Department of Labor's O*NET database, then looked at patterns across occupations and occupational categories.
Usage shifts since the launch
According to Anthropic, the share of usage rose modestly in several occupational categories, including coding, education and the sciences, along with healthcare applications. The company says computer and mathematical occupations showed the largest absolute increase in usage share, at 3 percentage points. Anthropic notes the coding increase was expected given Claude 3.7 Sonnet's improved coding benchmark scores, while it says the rise in other categories could reflect broader AI adoption, new applications of coding in those fields, or unexpected model capabilities.
Extended thinking mode usage
Anthropic reports that extended thinking mode, which lets the model reason longer on complex questions when activated by a user, was used predominantly in technical and creative contexts. The company says tasks associated with computer and information research scientists led usage at close to 10%, followed by software developers at around 8%, with multimedia artists at about 7% and video game designers at about 6%. Anthropic is releasing a dataset mapping each O*NET task to its associated thinking-mode usage on Hugging Face.
Augmentation versus automation
Anthropic says the overall balance between augmentative uses, such as learning or iterating with the model, and automative uses, such as having the model complete or debug a task directly, was largely unchanged from its first report, with augmentation accounting for 57% of usage. The company reports that learning interactions, where users ask Claude for information or explanations, rose from about 23% to about 28%. Anthropic says Community and Social Service tasks, including education and guidance counseling, approached 75% augmentation, while production and computer and mathematical occupations were closer to a 50-50 split; the company states no occupational category showed automation as dominant. Anthropic adds that copywriter and editor tasks showed the highest task iteration, where a person and the model co-write output, while translator and interpreter tasks showed among the highest directive behavior, where the model completes work with minimal human involvement. Librarians showed the highest share of learning interactions, at about 56%, and copywriters led task iteration at about 58%, according to the company.
New bottom-up usage taxonomy
Anthropic is also releasing what it calls a first-of-its-kind bottom-up taxonomy of Claude.ai usage, built with Clio from the same conversation dataset. It consists of 630 granular clusters organized into three hierarchy levels, each with descriptions, prevalence metrics and automation/augmentation breakdowns. Anthropic highlighted example clusters including help with water management systems, physics-based simulations, font selection, job application materials, battery technology guidance, and time zone handling in code. The company says the datasets behind both analyses are freely available to download, and that it plans to keep tracking these metrics as model capabilities advance. Anthropic also noted it is hiring for Societal Impacts Research Scientist, Research Engineer and Economist roles.
Source: Anthropic, "Anthropic Economic Index: Insights from Claude 3.7 Sonnet," published March 27, 2025.
Frequently asked questions
- What is the Anthropic Economic Index?
- Anthropic describes it as an initiative that regularly releases data and research on how AI affects labor markets and the economy over time.
- What data did this second report analyze?
- Anthropic analyzed 1 million anonymized Claude.ai Free and Pro conversations from the 11 days following the launch of Claude 3.7 Sonnet, using its Clio tool to map conversations to tasks in the U.S. Department of Labor's O*NET database.
- What did Anthropic find about extended thinking mode?
- Anthropic says extended thinking mode was used predominantly for technical and creative problem-solving tasks, led by computer and information research scientists at nearly 10% usage, followed by software developers at around 8%.
- Did the balance between AI augmentation and automation change?
- Anthropic reports the balance stayed essentially the same, with augmentation comprising 57% of usage, though learning-type interactions grew from about 23% to about 28% of the data.