2025 Posit Open Source Year in Review

2026-01-13

In 2025, our open-source teams focused on making R and Python faster, easier to use, and ready for AI. And it was quite a year! Key highlights include:

New tools like Air and duckplyr are speeding up data workflows and code formatting.
Updates to Quarto, Great Tables, and ggplot2 make it even easier to create polished, professional outputs.
New resources for building GenAI chatbots and evaluating the quality of LLM code.

We know many of you use open-source R and Python at work. Our professional product engineers also had an amazing 2025, building features that make enterprise research and data science safe and scalable. You can read their year in review here.

Posit and the Community

This year, Posit reaffirmed its commitment to a healthy ecosystem by joining the Open Source Pledge. We are proud to report a contribution of over $750,000 specifically toward the maintenance of the foundational tools the community relies on every day.
We gathered in Atlanta and online to engage, celebrate, and learn together at posit::conf(2025). If you missed a session, there are 2500+ hours of recordings available on YouTube, along with materials from our hands-on workshops. The conversation continues on our Discord server, which remained open after the event and has already grown to 1600+ community members.
Our Table and Plotnine contests encourage the community to push the boundaries of what’s possible in data communication. From the intricate clinical tables built with gt to the stunning Python visualizations in plotnine, this year’s entries prove the “grammar of graphics” is more vibrant than ever.

See the winners: Table Contest | Plotnine Contest

We launched a new podcast, The Test Set, designed to highlight the stories and strategies of leaders across the data field. You can read the recap here.

We continued our popular Data Science Hangout, hosting 45 sessions with thousands of unique attendees. Our DSH playlist videos on YouTube were watched more than 37,000 times last year!
We launched the Data Science Lab at the tail end of 2025, an unscripted, messy sharing session for technical data science topics. It’s been a blast, with nearly 500 registrants joining us to see how the work actually gets done.
Join us at future sessions: Data Science Hangout | Data Science Lab

Key Updates and New Releases

Speed, Concurrency, Performance, and Telemetry

In 2025, we released several tools to make your development workflow faster and your data outputs more stable.

Air: We introduced an extremely fast R formatter built on Rust.
Duckplyr: The dplyr backend powered by DuckDB fully joins the tidyverse! This allows you to work with billions of rows of data using the dplyr syntax you already know.
plumber2 v0.1.0: A complete reimagining of the web API framework for R, now featuring native support for promises and mirai for high-concurrency web services.

Performance was a headline for 2025, driven by the graduation of nanonext and mirai to the core r-lib family. These packages now form the backbone of parallel processing in R.

nanonext v1.7.0 and mirai v2.5.0: By leveraging NNG (Nanomsg Next Generation), these packages enable ultra-fast, asynchronous evaluation and cross-language interoperability.
purrr v1.1.0: The release of purrr 1.1.0 brought a game-changing feature: parallel processing, powered by mirai under the hood.
promises v1.5.0: The engine behind asynchronous Shiny got a major tune-up.

OpenTelemetry is an industry standard for collecting telemetry data (like traces and metrics) to understand how your code behaves in production. We have integrated it into a suite of packages to help you find bottlenecks and debug apps:

Shiny for R v1.12, mirai v2.5.0, promises v1.5.0, httr2 v1.2.2

Coming soon: OpenTelemetry support for ellmer and testthat! Learn more about it in this post on the Shiny blog.

Tidymodels and Machine Learning

sparsity support: tidymodels added full support for sparse data across the entire modeling workflow. his allows you to fit models on large, complex datasets faster and more memory-efficiently than before.
tune v2.0.0: This release overhauled parallel processing by moving toward future and mirai as backends.
tidypredict v1.0.0: Provides faster computations for tree-based models, more efficient tree representations, glmnet model support, and a change in how random forests are handled.
- With a massive update to the xgboost engine on CRAN, the tidymodels team worked hard to ensure a seamless transition for users, preserving existing workflows while unlocking new performance benefits.
Two New tidymodels Packages: November saw the release of filtro and important, two packages dedicated to supervised feature selection.
Orbital v0.3.0: Orbital translates tidymodels (R) and Scikit-learn (Python) pipelines directly into SQL. The R version now offers classification support, while the Python version features improved performance for tree-based models.

Reporting and Data Visualization

Quarto added additional features to help you turn your analyses into professional documents, websites, and presentations.

Quarto v1.8: This release introduced advanced “Brand” support. You can now define themes, logos, and typography in a single brand.yml file for your entire project.
- The quarto package v1.5 added features for streamlined workflows for R users.

Whether you’re an R enthusiast or a Pythonista, 2025 delivered many options for building world-class tables and visualizations.

gt v1.2.0 (R): Built in collaboration with GSK, this major release introduces new features for clinical reporting and better LaTeX and Word support.
Great Tables and pointblank (Python): Our Python-native table library, Great Tables, reached maturity this year. It now integrates with the new pointblank for Python to generate beautiful, interactive data validation reports.
- gt-extras: Introducing a new extension package for Great Tables users, as seen in this talk by Jules Walzer-Goldfeld at PyData Boston.
ggplot2 v4.0.0 (R): This milestone release moved to the S7 object-oriented system for more robust extensions. It also added new theme arguments for setting default geoms and palettes.
Plotnine v0.15 (Python): You can now combine multiple independent plots into a single layout (similar to patchwork in R) using the grammar of graphics in Python.

All About AI

What would a 2025 update be without AI?

ellmer (R) and chatlas (Python): Our key packages for chatting with LLMs support streaming, tool calling, and structured data extraction from OpenAI, Claude, and Gemini.
We released a host of updates to our suite of AI packages, including btw, mcptools, ragnar, and vitals.
side::kick() is a new experimental coding agent for RStudio users built entirely in R.

The 2025 updates for Shiny focused on helping you integrate apps with AI.

New Shiny for Python AI documentation: For our Python users, we’ve added comprehensive guides to building GenAI chatbots. They cover topics such as streaming responses to testing different models with a playground template.
shinychat: Available for both R and Python, shinychat makes it easy to create Shiny AI applications with streaming content, tool call displays, bookmarking, and more.
querychat: Available for R and Python, querychat allows you to use large language models to query tabular data.

Veerle Eeftink – van Leemput wrote a series of articles on building real, useful LLM-powered apps without getting buried in jargon: Part 1 | Part 2 | Part 3

Data science is being transformed by large language models, and Posit is providing the toolkit to lead that charge safely and effectively with Responsible AI.

While the hype is high, our recent research shows that local models are not yet advanced enough to power coding or data science agents.
We’ve released new resources to help you measure how well AI writes R code, and a guide on privacy and LLM assistants to help enterprises evaluate different providers for privacy and security.

Follow the AI Newsletter to stay up to speed with what’s happening in the AI world, both at Posit and beyond.

Looking forward to 2026

In 2026, we will continue to build high-performance, open-source tools for R and Python that are ready for the next wave of data science. We would be delighted to show you what we’re working on at the upcoming posit::conf(2026), happening in Houston, TX, on September 14-16!

Thank you for being part of this community. We can’t wait to see what you create next!