Wolfram Language & Mathematica 14.3: Major Update with AI, Dark Mode

Wolfram Research has unveiled Version 14.3 of the Wolfram Language and Mathematica, a substantial update that introduces a wide array of new functionalities and refinements across its computational landscape. Despite its minor version designation, this release delivers significant advancements, many of which address long-standing user requests, reinforcing the platform’s commitment to design consistency and backward compatibility over nearly four decades.

A prominent new feature is the comprehensive Dark Mode support. Beyond a simple inversion of colors, Version 14.3 intelligently adapts thousands of user interface elements and graphics to ensure optimal aesthetics and readability in dark environments. For instance, plots automatically adjust axes colors while preserving data curve colors, which were pre-optimized in Version 14.2 for both light and dark display. The system introduces LightDarkSwitched for explicit color control, LightDarkAutoColorRules for global automatic switching, and SystemColor to align with operating system themes. Textual content and theme-based colors, accessible via ThemeColor, also dynamically adapt, showcasing a deep algorithmic approach to user interface aesthetics.

The release further deepens the integration with modern Artificial Intelligence. Positioned as a “universal computational agent” or “tool” for Large Language Models (LLMs), the Wolfram Language leverages its precise computational capabilities to complement the heuristic nature of LLMs. Version 14.3 introduces LLMGraph, enabling users to define complex “agentic workflows” directly within the Wolfram Language. These graphs can orchestrate parallel execution of LLM prompts and Wolfram Language code, complete with test functions to guide decision-making, facilitating sophisticated AI-driven processes.

Data Fitting and Analysis receive a significant boost with ListFitPlot, a new function for easily visualizing fits to data, including local, linear, and exponential models. Complementing existing tools, LocalModelFit and KernelModelFit offer new non-parametric fitting options. The PlotFit option allows adding fits to existing plot types, and ListFitPlot3D extends fitting capabilities to 3D surfaces.

Mapping capabilities are visually enhanced with updated styles and rendering, including full dark mode support. Maps now default to resolution-independent vector graphics for crisp display at any zoom level. New features like night-time satellite imagery and the ability to generate bland base maps (for better data overlay) improve customization. GeoReposition allows precise manipulation of geographic objects, demonstrating concepts like continental drift or the true scale of landmasses, even off-planet.

Graphics and Color improvements extend to a new concept of “standard colors” like StandardRed and DarkRed, offering a more “designed” aesthetic than pure RGB values, optimized for both light and dark modes. The color picker has been completely redesigned for intuitive use, incorporating a color wheel and streamlined light/dark switching. Default colors across various plotting functions, such as DensityPlot and ArrayPlot, have been “spiffed up” for greater visual impact. LabelingTarget provides fine-grained control over plot labels to prevent overlap, and PlotInteractivity allows disabling interactive elements for efficiency or print optimization.

Significant advancements have been made in Non-commutative Algebra. The long-standing NonCommutativeMultiply (typed as ** and now rendered as ⦻) finally gains computational power. New functions like NonCommutativeExpand allow for canonical form expansion, and the generalization of Gröbner bases to the non-commutative case enables powerful symbolic array operations and applications in physics and functional programming. The release also introduces GeneralizedPower and Commutator for non-commutative expressions.

The realm of Matrices sees an expansion with RangeSpace complementing NullSpace, and Projection now supporting projection onto subspaces. Four new matrix decompositions are introduced, including EigenvalueDecomposition and FrobeniusDecomposition (offering robust diagonalization). New functions like MatrixMinimalPolynomial and MatrixPolynomialValue extend polynomial operations to matrices, while polynomial HermiteDecomposition and reduction functions enhance algebraic manipulation.

In Geometric Computation, SurfaceContourPlot3D allows plotting functions directly on 3D surfaces, and HighlightRegion enables visual annotation of specific areas on regions in any dimension. New curvature computations (GaussianCurvature, mean, max/min) provide precise analysis of surface geometry. FindShortestCurve and ShortestCurveDistance facilitate geodesic calculations and path planning on arbitrary regions, including complex meshes and for obstacle avoidance in robotics. SubdivisionRegion offers a powerful method for generating smooth, realistic geometry from coarse approximations, while SmoothMesh, SimplifyMesh, and Remesh provide tools for refining and repairing 3D meshes.

Chemistry and Biosciences see enhanced molecular visualization with property-based coloring for atoms and bonds. MoleculeFeatureDistance offers a quantitative measure of molecular similarity, enabling applications like molecular clustering. MoleculeModify is expanded to include inversion of molecular structure. A major new capability is local Protein Folding, allowing users to perform machine-learning-based protein structure predictions on their own machines, bypassing external API limitations. BioMoleculeAlign assists in comparing predicted structures with experimental data, and new measures like Ramachandran angles provide deeper insights into protein shape.

For Engineering and Control Systems, Version 14.3 introduces SystemModelValidate for verifying system behavior against specifications. Using temporal logic constructs like SystemModelAlways, SystemModelEventually, and SystemModelSustain, engineers can define conditions (e.g., battery temperature limits) and assess if a system model meets them, identifying failure points. Control system workflows are streamlined with interactive model manipulation and new analytical tools like PoleZeroPlot for stability analysis and LQ controller design.

Several Programming Language Enhancements improve developer experience. The long-requested multi-argument With construct simplifies nested variable scoping. The new Cyclic function provides a convenient way to treat lists as infinitely repeating sequences, useful for both computation and cyclic styling.

Tabular Data capabilities, introduced in Version 14.2, are further rounded out. New import sources include OneDrive, Kaggle, and direct in-core import from relational databases (SQLite, Postgres, MySQL, SQL Server, Oracle) via DataConnectionObject. Support for Apache, Common, and Extended Log files, as well as JSON Lines, has been added. New functions like ColumnwiseCombine and enhanced JoinAcross provide powerful ways to combine data from multiple Tabular objects with flexible criteria. Tabular data Styling is also introduced, allowing programmatic control over display size, appearance elements, backgrounds (including conditional styling based on values), and item styles.

In Semantic Text Processing, SemanticRanking offers a new function to rank choices based on textual descriptions, leveraging modern language models. It complements SemanticSearch by providing detailed relevance scores for smaller sets of options. An enhanced, pre-trained FeatureExtract for text further improves classification and prediction tasks.

A core new capability for the Wolfram Language compiler is Compiled Functions that can Pause and Resume. IncrementalFunction enables the creation of coroutines and generators using IncrementalYield and IncrementalReceive. This allows for incremental computation, maintaining internal state automatically, which is particularly useful for enumerating large datasets that wouldn’t fit in memory or for processing data streams.

External Computation sees significant performance and usability improvements. Python runtime provisioning is dramatically faster, including for specific dependencies. A highly streamlined R integration is introduced, allowing multiple R sessions within a single Wolfram Language session, each with independent dependencies, and supporting ExternalFunction for seamless R code execution.

For Notebooks into Presentations, Version 14.3 solves the long-standing aspect ratio challenge. A new full-slide image template in Presenter Notebooks automatically scales images and graphics to fit or fill the screen, ensuring professional-looking slideshows without manual adjustments.

User Interface Polishing continues with subtle but impactful refinements, such as smarter handling of character sequences like -> for arrow keys and single-character delimiter wrapping (e.g., typing { around a selection automatically wraps it with {...}). Support for separate British and American English spelling dictionaries is also added.

Finally, Markdown is now a first-class format, allowing easy import and export of Markdown files and fragments, facilitating communication with other systems and LLMs. The web documentation also receives a new, highly functional navigation sidebar for quick access to options and sections within function pages.

Beyond these highlights, Version 14.3 includes numerous other enhancements across video processing (stabilization, object tracking), image import (.avif), local speech synthesis with neural net models, refined date handling, more efficient RandomTree generation, multithreading for BitVector data structures, GPU support for additional functions, axisymmetric fluid flow solving in PDE, new biochemistry connections (UniProt, AlphaFold), and compiler introspection tools.

Version 14.3 of Wolfram Language and Mathematica is available for download for desktop systems and is live in the Wolfram Cloud, representing the latest strides in the platform’s continuous research and development efforts.