2026 DX benchmarks are now available

Industry benchmarks help put metrics into context by showing how your engineering organization compares to similar companies. At DX, we’ve assembled the world’s largest body of engineering benchmark data, consisting of millions of data points across sectors and geographies.

DX customers can now access the updated 2026 benchmark dataset. This year’s release introduces new geography-based segments, more granular company size cohorts, expanded APAC-focused variants, a new PR revert rate benchmark, and refreshed benchmark data, giving you more relevant comparisons and a current view of engineering performance.

What’s new in the 2026 benchmarks

This year’s update expands benchmark coverage to help you compare against organizations that more closely match your own. It also introduces a new benchmark metric for measuring code quality.

Consistent with our 2025 benchmarks, each new segment includes at least 30,000 individuals and/or 200 organizations, based on data from DX customers as well as research panel data. As our customer base grows, we’re able to offer increased granularity across segments.

New benchmark segments:

New geographic cohorts by HQ: Europe, Silicon Valley, Latin America, APAC
New APAC contractors cohort
More granular company sizes: <100*, 100–500, 500–1000, 1,000–2,500, and 2,500+ engineers
Expanded APAC variants across benchmark segments
New large enterprise technology cohort: Tech (2500+ engineers)
New fintech cohort: Fintech (<100 engineers)

New system metric benchmarked:

PR revert rate: a benchmark for understanding how frequently pull requests are reverted after merge

Shifts from 2025 to 2026

Beyond the new benchmark segments and metrics, the updated dataset highlights several notable shifts in engineering performance from 2025 to 2026.

Core 4 metrics

Overall, the 2026 Core 4 benchmarks point to improvements in speed, effectiveness, and quality, while impact declined slightly.

Developer Experience Index (DXI) (self-reported) increased by 2 points, indicating modest improvements in developers’ self-reported experience.

PR Throughput (systems-based) increased by nearly 20%, with noticeable increases in median organizational output following the release of Anthropic’s Opus 4.6 model in February.

Defect ratio (systems-based) improved by 16.9%, suggesting software quality has remained stable as AI coding tools have become more widely adopted, though additional research is needed.

Innovation ratio (systems-based) decreased by 10%, potentially reflecting a greater share of engineering effort spent on maintenance, technical debt, and refactoring.

Driver-level trends

Beyond the Core 4 metrics, the updated benchmark data highlights several notable trends across the individual drivers.

Code maintainability improved across nearly every benchmark segment, with Java engineering organizations seeing the largest gains (+15). Only non-tech organizations with 100–500 engineers experienced a slight decline (-1). One possible explanation is that AI is helping engineers better understand and navigate complex codebases, even as the volume of code continues to grow.

Customer focus increased across every benchmark segment. One possible explanation is that AI is reducing the time spent on KTLO work, giving developers more capacity to build customer-facing features.

Documentation improved across every benchmark segment, with the largest increase among Finserv organizations with 250+ engineers (+19). This is consistent with what we’ve observed across our customer base over the past year. As AI becomes part of developers’ daily workflows, it’s becoming easier to create, update, and find internal documentation.

Several workflow-related metrics moved in the opposite direction. Incident handling, review turnaround, and incremental delivery all declined across most benchmark segments.

The slowing of review turnaround is consistent with what we’ve observed across many customer organizations. AI is generating more code, increasing review volume faster than many teams have adapted their review processes.

Incremental delivery may be slowing for the same reason. As review volume increases, work takes longer to move through the delivery pipeline. AI is helping teams generate code more efficiently, but many organizations are still adapting the workflows around code review and delivery.

Production debugging was one of the few drivers to improve, with gains across nearly every benchmark segment.

Taken together, these findings suggest that AI is helping teams produce and maintain code more effectively, but many organizations are still adapting the workflows around code review, delivery, and operations. As AI adoption matures, we expect these processes to continue evolving.

Workflow metrics

Many of the largest year-over-year changes appeared in workflow-related metrics. Given how quickly AI adoption and proficiency increased over the past year, these shifts are largely consistent with what we’d expect as AI becomes part of developers’ day-to-day workflows.

Compared to the 2025 benchmarks:

AI-authored code (self-reported) increased by 140%.
AI time savings (self-reported) increased by 60%, from 3.9 to 6.2 hours per developer per week.

These changes provide important context for the benchmark shifts outlined above. As AI becomes more deeply embedded in software development, we’re seeing meaningful improvements in code generation and developer productivity alongside new bottlenecks in review and delivery workflows.

For a deeper look at these trends and the underlying research, read our Q2 2026 AI Impact Report.

Getting started

The 2026 benchmarks are now available to all DX customers.

To move to the 2026 industry benchmarks, set them as your organization’s default in Admin, or reach out to your DX representative. The 2025 benchmarks will remain available if you prefer to continue using them.