Anthropic, the developer of the Claude AI model, has released its fourth Economic Index report, analyzing real-world usage to gauge how artificial intelligence is transforming work productivity and job structures. Drawing from conversations in November 2025 primarily involving the Claude Sonnet 4.5 model, the study introduces five core metrics, or economic primitives, to track AI’s effects: task complexity, required skill levels, intended purpose such as work or education, the degree of AI independence, and overall success rates. These insights come from a privacy-focused method that examines interactions on the consumer-facing Claude.ai platform and the company’s API, which serves mostly business clients.
The report addresses key uncertainties about AI’s role in the workplace, including whether it accelerates tasks, which activities it handles most effectively, and its potential to reshape careers. Previous Anthropic analyses have explored AI adoption by job type and salary, focused on software engineering, and mapped usage patterns across nations and U.S. states. This latest edition builds on those efforts by applying the new primitives to delve into task-level dynamics and broader economic implications.
Findings reveal that Claude provides the greatest time savings for more intricate tasks. On Claude.ai, activities needing a high school-level understanding see a ninefold speedup, while those requiring a college education achieve a 12-fold increase. API users experience even larger gains. This pattern suggests AI benefits are flowing toward roles demanding substantial expertise, aligning with data showing white-collar workers leading AI adoption.
Even factoring in success probabilities, the trend persists. Claude completes college-level tasks successfully 66 percent of the time, compared to 70 percent for those below high school level, but the productivity boost still rises more with complexity than success rates decline.
Regarding task duration, the report complements benchmarks from the nonprofit METR, which test AI on extended activities. While METR indicates Claude Sonnet 4.5 reaches 50 percent success on two-hour tasks, Anthropic’s data shows 50 percent reliability for 3.5-hour efforts via API and up to 19 hours on Claude.ai. Differences stem from users breaking down projects iteratively and selecting feasible challenges, unlike standardized tests.
Usage patterns differ by national wealth. In higher-income countries, Claude supports work and personal activities more often, while lower-income nations emphasize educational applications. This mirrors a progression where AI starts with learning in developing economies and expands to diverse uses as prosperity grows. Anthropic highlighted its partnership with Rwanda’s government and training provider ALX to foster AI skills and provide extended access to premium features, easing the shift from study to professional applications.
On job coverage, pooled data across reports indicates 49 percent of occupations involve Claude in at least a quarter of tasks, up from 36 percent in January 2025. Adjusting for success and task duration alters priorities: roles like data entry and radiology face heavier AI influence than simple coverage suggests, while teaching and programming see less. AI tends to tackle higher-skill elements within jobs, averaging 14.4 years of education needed versus the overall 13.2-year norm, potentially simplifying roles if these duties automate, though labor markets could adapt.
At a macroeconomic level, incorporating success rates tempers earlier projections. Without adjustments, AI could add 1.8 percentage points to annual U.S. labor productivity growth over the next decade, roughly doubling recent trends. Accounting for reliability, estimates drop to 1.2 points for Claude.ai tasks and 1.0 for API ones, still a substantial lift reminiscent of the late 1990s tech boom. Future model advances, like the newer Claude Opus 4.5, could amplify these gains through broader or more refined applications.
Updates on ongoing trends show persistent concentration: the top 10 of 3,000 work tasks claim 24 percent of Claude.ai activity, up slightly from earlier. Computer and math-related work dominates, comprising a third of consumer sessions and nearly half of API ones. Interaction styles have shifted, with augmentation now edging out automation at 52 percent to 45 percent on Claude.ai, though automation’s share grows gradually over time.
Geographically, leading users remain the United States, India, Japan, the United Kingdom, and South Korea, tied to economic strength. Within the U.S., adoption is spreading more evenly across states, potentially uniform nationwide in two to five years if patterns hold.
Overall, the analysis underscores AI’s uneven rollout, clustered in select regions, professions, and functions. As Claude evolves, Anthropic anticipates tougher assignments, higher reliability, and migration toward business tools, signaling deeper workforce shifts. The full report is available on Anthropic’s research page, offering data for further study by policymakers and experts navigating this technological wave.
