Wolfram Language and Mathematica version 15 adds LLM functions and performance boosts

Wolfram Research released Wolfram Language and Mathematica version 15 on June 16, 2026, bundling a suite of AI‑driven primitives and a refreshed core library [Hacker News].

── What shipped ──

The launch post lists more than 70 new functions, most centered on large‑language‑model (LLM) access. The flagship addition is LLMFunction, which lets users call any hosted LLM—including Wolfram’s own model—directly from Wolfram code and receive structured results for downstream symbolic pipelines. A companion TextToImage primitive wraps a fine‑tuned StableDiffusion model, exposing image generation as a one‑line call.

Version 15 also expands the NeuralNetworkRepository by 250 models and introduces SymbolicSimplifyAI, a hybrid symbolic‑numeric routine that uses an LLM to suggest simplifications for otherwise intractable expressions. Data‑science workflows gain TimeSeriesForecastAI, which automatically selects the best forecasting architecture (ARIMA, Prophet, or transformer) based on the input series.

Performance improvements are concrete: the kernel’s symbolic integrator runs about 20 % faster on benchmark suites involving nested radicals, and the new GPUCompiler adds support for AMD RDNA 3 GPUs, cutting matrix‑multiply latency by roughly 15 %. The cloud‑deployment model also received a streamlined packaging format that halves container start‑up time [Hacker News].

── Why it matters ──

Built‑in LLM access removes the need for external API integrations, keeping the entire stack inside the Wolfram ecosystem. The hybrid symbolic‑AI pipeline lets researchers offload complex simplifications—such as high‑order differential equations—to SymbolicSimplifyAI, accelerating prototype cycles. Speed gains in symbolic integration translate directly to shorter runtimes for physics and engineering simulations, preserving Wolfram’s competitiveness for large‑scale workloads. Finally, the enlarged model repository turns the Wolfram Language into a one‑stop shop for pretrained AI, eliminating separate model‑hosting infrastructure.

── Editor's take ──

Wolfram’s aggressive AI integration is a double‑edged sword. Existing customers gain a powerful, turnkey solution, but the move deepens lock‑in to a proprietary stack at a time when the open‑source community is coalescing around Python‑centric AI toolchains. Engineers who value flexibility may find the trade‑off worth debating, especially as Wolfram Cloud credit costs rise.

Poll

Which AI‑enabled computational platform do you trust for production workloads?

• Wolfram Language 15
• Python with LangChain
• Node.js with OpenAI SDK
• Julia with Flux