{"success":true,"data":{"author":"Anthropic Economic Research","bucketAccessType":"public","bucketName":"Anthropic/EconomicIndex","bucketPrefix":"","category":"Machine Learning & AI","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:20.253Z","description":"Data releases behind the Anthropic Economic Index, which measures how AI is being used across real tasks in the economy by mapping privacy-preserving aggregates of Claude conversations onto the US Department of Labor O*NET task and occupation taxonomy. The repository holds one folder per release: the February 2025 initial release (O*NET task mappings, automation versus augmentation shares, occupational category and wage distributions with BLS employment data), the March 2025 release (Claude 3.7 Sonnet data, cluster-level data and extended thinking fractions), the September 2025 release (geographic usage by country and US state and first-party API data, with GDP, population and Census inputs and analysis code), the January 2026 release (economic primitives), March 2026 and June 2026 releases, plus labor_market_impacts (job exposure and task penetration) and robot_exposure folders. Files are CSV, TSV and Excel inputs with Jupyter notebooks and Python preprocessing scripts; the repository uses about 0.9 GB (measured 2026-11-14). Open with pandas. Uses include labour economics, AI exposure and adoption measurement, occupational task analysis and regional diffusion studies. The card states the data are released under CC-BY and the code under MIT; cite the relevant Economic Index report.","displayTitle":"Anthropic Economic Index","id":"22ce1e36-8381-53c9-82d1-a8c444eeba0f","license":"CC-BY-4.0","licenseEvidenceUrl":"https://huggingface.co/datasets/Anthropic/EconomicIndex","objectCountEstimate":86,"probeStatus":"ok","referenceUrl":"https://arxiv.org/abs/2503.04761","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"Anthropic/EconomicIndex\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"Anthropic/EconomicIndex\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=Anthropic, license:mit, downloads=23785"],"format":"mixed","generatedAt":"2026-11-14T14:15:32.590Z","layout":{"example":"labor_market_impacts/job_exposure.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"occ_code,title,observed_exposure\n11-1011,Chief Executives,0.0333\n11-1021,General and Operations Managers,0.1378\n11-1031,Legislators,0.0\n11-2011,Advertising and Promotions Managers,0.1731\n11-2021,Marketing Managers,0.3195\n11-2022,Sales Managers,0.0433\n11-2032,Public Relations Managers,0.2315\n11-3012,Administrative Services Managers,0.0\n11-3013,Facilities Managers,0.1345\n11-3021,Computer and Information Systems Managers,0.1559\n11-3031,Financial Managers,0.3907\n11-3051,Industrial Production Managers,0.0132\n11-","encoding":"utf8","key":"labor_market_impacts/job_exposure.csv","truncatedAt":512}],"sampleObjects":["labor_market_impacts/job_exposure.csv","labor_market_impacts/task_penetration.csv","release_2025_02_10/SOC_Structure.csv","release_2025_02_10/automation_vs_augmentation.csv","release_2025_02_10/bls_employment_may_2023.csv"],"sidecars":[{"bytes":"---\nlanguage: en\npretty_name: EconomicIndex\ntags:\n- AI\n- LLM\n- Economic Impacts\n- Anthropic\nlicense: mit\nviewer: true\nconfigs:\n- config_name: release_2026_01_15\n  data_files:\n  - split: raw_claude_ai\n    path: \"release_2026_01_15/data/intermediate/aei_raw_claude_ai_2025-11-13_to_2025-11-20.csv\"\n  - split: raw_1p_api\n    path: \"release_2025_09_15/data/intermediate/aei_raw_1p_api_2025-11-13_to_2025-11-20.csv\"\n---\n\n\n# The Anthropic Economic Index\n\n## Overview\n\nThe Anthropic Economic Index provides insights into how AI is being incorporated into real-world tasks across the modern economy.\n\n## Data Releases\n\nThis repository contains multiple data releases, each with its own documentation:\n\n- **[Labor market impacts](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/labor_market_impacts)**: Job exposure and task penetration data\n- **[Robot Exposure](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/robot_exposure)**: Job and task exposure to robots\n- **[2026-06-26 Release](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/release_2026_06_26)**: Updated analysis with Artifacts and monthly aggregates\n- **[2026-03-24 Release](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/release_2026_03_24)**: Updated analysis with Opus 4.5/4.6 and learning curves\n- **[2026-01-15 Release](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/release_2026_01_15)**: Updated analysis with economic primitives and Sonnet 4.5\n- **[2025-09-15 Release](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/release_2025_09_15)**: Updated analysis with geographic and first-party API data using Sonnet 4\n- **[2025-03-27 Release](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/release_2025_03_27)**: Updated analysis with Claude 3.7 Sonnet data and cluster-level insights\n- **[2025-02-10 Release](https://huggingface.co/datasets/Anthropic/EconomicIndex/tree/main/release_2025_02_10)**: Initial release with O*NET task mappings, automation vs. augmentation data, and more\n\n\n## Resources\n\n- [Index Home Page](https://www.anthropic.com/economic-index)\n- [6th report](https://www.anthropic.com/research/economic-index-june-2026-report)\n- [5th report](https://www.anthropic.com/research/economic-index-march-2026-report)\n- [4th report](https://www.anthropic.com/research/anthropic-economic-index-january-2026-report)\n- [3rd report](https://www.anthropic.com/research/anthropic-economic-index-september-2025-report)\n- [2nd report](https://www.anthropic.com/news/anthropic-economic-index-insights-from-claude-sonnet-3-7)\n- [1st report](https://www.anthropic.com/news/the-anthropic-economic-index)\n\n- [What work can robots do?](https://www.anthropic.com/research/what-work-can-robots-do)\n- [Labor market impacts](https://www.anthropic.com/research/labor-market-impacts)\n\n## License\n\nData released under CC-BY, code released under MIT License.\n\n## Contact\n\nFor press inquiries, contact press@anthropic.com. For all other questions, reach out to econ-research@anthropic.com.\n\n## Citation\n\n### Sixth release\n```\n@online{anthropic2026aeiv6,\n        author = {Maxim Massenkoff and Eva Lyubich and Szymon Sacher and Zoe Hitzig and Shaoyi Zhang and Ryan Heller and Peter McCrory},\n        title = {Anthropic Economic Index report: Cadences},\n        date = {2026-06-26},\n        year = {2026},\n        url = {https://www.anthropic.com/research/economic-index-june-2026-report},\n}\n\n### Fifth release\n\n```\n@online{anthropic2026aeiv5,\n        author = {Maxim Massenkoff and Eva Lyubich and Peter McCrory and Ruth Appel and Ryan Heller},\n        title = {Anthropic Economic Index report: Learning curves},\n        date = {2026-03-24},\n        year = {2026},\n        url = {https://www.anthropic.com/research/economic-index-march-2026-report},\n}\n```\n\n### Fourth release\n\n```\n@online{anthropic2026aeiv4,\n        author = {Ruth Appel and Maxim Massenkoff and Peter McCrory and Miles McCain and Ryan Heller and Tyler Neylon and Alex Tamkin},\n        title = {Anthropic Economic Index report: economic primitives},\n      ","encoding":"utf8","key":"README.md","name":"README.md","size":5262,"truncated":true},{"bytes":"---\nlicense: mit\npretty_name: EconomicIndex\ntags:\n- text\nviewer: true\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: \"onet_task_mappings.csv\"\n---\n## Overview\nThis directory contains O*NET task mapping and automation vs. augmentation data from \"Which Economic Tasks are Performed with AI? Evidence from Millions of Claude Conversations.\" The data and provided analysis are described below.\n\n**Please see our [blog post](https://www.anthropic.com/news/the-anthropic-economic-index) and [paper](https://assets.anthropic.com/m/2e23255f1e84ca97/original/Economic_Tasks_AI_Paper.pdf) for further visualizations and complete analysis.**\n\n## Data\n\n- `SOC_Structure.csv` - Standard Occupational Classification (SOC) system hierarchy from the U.S. Department of Labor O*NET database\n- `automation_vs_augmentation.csv` - Data on automation vs augmentation patterns, with columns:\n  - interaction_type: Type of human-AI interaction (directive, feedback loop, task iteration, learning, validation)\n  - pct: Percentage of conversations showing this interaction pattern\n  Data obtained using Clio (Tamkin et al. 2024)\n- `bls_employment_may_2023.csv` - Employment statistics from U.S. Bureau of Labor Statistics, May 2023\n- `onet_task_mappings.csv` - Mappings between tasks and O*NET categories, with columns:\n  - task_name: Task description\n  - pct: Percentage of conversations involving this task\n  Data obtained using Clio (Tamkin et al. 2024)\n- `onet_task_statements.csv` - Task descriptions and metadata from the U.S. Department of Labor O*NET database\n- `wage_data.csv` - Occupational wage data scraped from O*NET website using open source tools from https://github.com/adamkq/onet-dataviz\n\n## Analysis\n\nThe `plots.ipynb` notebook provides visualizations and analysis including:\n\n### Task Analysis\n- Top tasks by percentage of conversations\n- Task distribution across occupational categories\n- Comparison with BLS employment data\n\n### Occupational Analysis  \n- Top occupations by conversation percentage\n- Occupational category distributions\n- Occupational category distributions compared to BLS employment data\n\n### Wage Analysis\n- Occupational usage by wage\n\n### Automation vs Augmentation Analysis\n- Distribution across interaction modes\n\n## Usage\nTo generate the analysis:\n\n1. Ensure all data files are present in this directory\n2. Open `plots.ipynb` in Jupyter\n3. Run all cells to generate visualizations\n4. Plots will be saved to the notebook and can be exported\n\nThe notebook uses pandas for data manipulation and seaborn/matplotlib for visualization. Example outputs are contained in the `plots\\` folder.\n\n**Data released under CC-BY, code released under MIT License**\n\n## Contact\nYou can submit inquires to kunal@anthropic.com or atamkin@anthropic.com. We invite researchers to provide input on potential future data releases using [this form](https://docs.google.com/forms/d/e/1FAIpQLSfDEdY-mT5lcXPaDSv-0Ci1rSXGlbIJierxkUbNB7_07-kddw/viewform?usp=dialog).","encoding":"utf8","key":"release_2025_02_10/README.md","name":"README.md","size":2981,"truncated":false},{"bytes":"# Anthropic Economic Index: Insights from Claude 3.7 Sonnet\n## Analysis Replication Notebook\n\nThis notebook contains the code used to produce the visualizations and analysis for the Anthropic Economic Index report based on Claude 3.7 Sonnet data. It analyzes how different occupations interact with AI systems through automation and augmentation patterns derived from real-world usage data.\n\n## Data Files in this Directory\n\n- **cluster_level_dataset**: A folder containing data released at the cluster level, including mappings to O*NET tasks, automation vs. augmentation, and \"extended thinking\" mode fraction\n- **onet_task_statements.csv**: Contains O*NET task statements with their associated occupational codes\n- **SOC_Structure.csv**: Standard Occupational Classification (SOC) structure data with major group codes and titles\n- **task_pct_v1.csv**: Version 1 of task percentage data\n- **task_pct_v2.csv**: Version 2 of task percentage data (current)\n- **automation_vs_augmentation_by_task.csv**: Data on automation vs. augmentation classifications by task\n- **automation_vs_augmentation_v1.csv**: Version 1 of automation vs. augmentation interaction type data\n- **automation_vs_augmentation_v2.csv**: Version 2 of automation vs. augmentation interaction type data\n- **task_thinking_fractions.csv**: Fraction of each O*NET task with its associated \"extended thinking\" mode fraction\n\n## Data Dictionary\n\n### onet_task_statements.csv\n| Field | Description |\n|-------|-------------|\n| O*NET-SOC Code | Occupational code from the O*NET-SOC system |\n| Title | Occupational title |\n| Task | Description of specific occupational task |\n\n### SOC_Structure.csv\n| Field | Description |\n|-------|-------------|\n| Major Group | Two-digit code identifying major occupational group |\n| SOC or O*NET-SOC 2019 Title | Title of the major occupational group |\n\n### task_pct_v1.csv and task_pct_v2.csv\n| Field | Description |\n|-------|-------------|\n| task_name | Normalized name of the task |\n| pct | Percentage of task prevalence in dataset |\n\n### automation_vs_augmentation_by_task.csv\n| Field | Description |\n|-------|-------------|\n| task_name | Normalized name of the task |\n| directive | Ratio indicating directive automation pattern (0-1) |\n| feedback_loop | Ratio indicating feedback loop automation pattern (0-1) |\n| validation | Ratio indicating validation augmentation pattern (0-1) |\n| task_iteration | Ratio indicating task iteration augmentation pattern (0-1) |\n| learning | Ratio indicating learning augmentation pattern (0-1) |\n| filtered | Ratio indicating filtered (excluded) tasks (0-1) |\n\n### automation_vs_augmentation_v1.csv and automation_vs_augmentation_v2.csv\n| Field | Description |\n|-------|-------------|\n| interaction_type | Type of interaction (directive, feedback loop, validation, task iteration, learning, none) |\n| pct | Percentage of this interaction type in the dataset |\n\n### task_thinking_fractions.csv\n| Field | Description |\n|-------|-------------|\n| task_name | Normalized name of the task |\n| thinking_fraction | Ratio of this task that used extended thinking mode |\n\n## Running the analysis\nOpen `v2_report_replication.ipynb` in a notebook editor and run the cells in order.","encoding":"utf8","key":"release_2025_03_27/README.md","name":"README.md","size":3205,"truncated":false},{"bytes":"# Cluster Level Data\n\nThis folder contains cluster level data for the second Economic Index release associated with Claude 3.7 Sonnet Data. It contains hierarchical cluster descriptions, as well as associated prevalence metrics for each cluster (% of records, % of users). It also includes mappings to [O*NET Tasks](https://www.onetonline.org/), collaboration pattern ratios, and ratios associated with whether or not Claude Sonnet 3.7's \"Thinking\" feature was used during the conversation.\n\n## Files in this Directory\n\n- **cluster_level_dataset.tsv**: Tab-separated values file containing the cluster data with all fields described in the data dictionary below. This is the primary dataset file for analysis.\n\n- **cluster_level_example_analysis.ipynb**: Jupyter notebook demonstrating example analyses you can perform with the cluster level dataset. This notebook includes code for loading the data, basic exploratory analysis, and visualization techniques to help understand the cluster patterns and their relationships to O*NET tasks.\n\n## Data Dictionary\n\n| Field | Description |\n|-------|-------------|\n| cluster_name_0 | Name of the level 0 (most granular) cluster |\n| cluster_description_0 | Detailed description of the level 0 cluster |\n| cluster_name_1 | Name of the level 1 (intermediate) cluster |\n| cluster_description_1 | Detailed description of the level 1 cluster |\n| cluster_name_2 | Name of the level 2 (broadest) cluster |\n| cluster_description_2 | Detailed description of the level 2 cluster |\n| percent_records | Percentage of total records that belong to this Level 0 cluster |\n| percent_users | Percentage of total users who have used this Level 0 cluster |\n| onet_task | Description of the associated O*NET task |\n| collaboration:directive_ratio | Ratio of conversations with directive collaboration patterns |\n| collaboration:feedback loop_ratio | Ratio of conversations with feedback loop collaboration patterns |\n| collaboration:learning_ratio | Ratio of conversations with learning collaboration patterns |\n| collaboration:none_ratio | Ratio of conversations with no collaboration patterns |\n| collaboration:task iteration_ratio | Ratio of conversations with task iteration collaboration patterns |\n| collaboration:validation_ratio | Ratio of conversations with validation collaboration patterns |\n| has_thinking_ratio | Ratio of conversations where the \"Thinking\" feature was used |\n\n## Bucketing Adjustment\n\nFor percent_records and percent_users fields, we applied a bucketing adjustment to enhance privacy while preserving the overall distribution:\n\n1. Clusters were sorted by their prevalence metrics (percent_records and percent_users).\n2. 100 buckets were created.\n3. Clusters were assigned to buckets and the average prevalence within each bucket was calculated.\n4. The original values were replaced with the bucket averages to reduce precision while maintaining the distribution.\n\n\n\n","encoding":"utf8","key":"release_2025_03_27/cluster_level_data/README.md","name":"README.md","size":2917,"truncated":false}],"sourceObjectKey":"labor_market_impacts/job_exposure.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/Anthropic/EconomicIndex/labor_market_impacts/job_exposure.csv\")\ndf.head()","starterSamplePath":"labor_market_impacts/job_exposure.csv","totalObjects":86},"summary":"Data releases behind the Anthropic Economic Index, which measures how AI is being used across real tasks in the economy by mapping privacy-preserving aggregates of Claude…","title":"Anthropic Economic Index","updatedAt":"2026-11-14T17:51:15.172Z","visibility":"public","websiteUrl":"https://www.anthropic.com/economic-index"}}