Modular grants fund Mojo RL, physics, FFT, JSON and HTTP projects

Modular published the cohort of its Community Grant Program on October 9, 2026, naming six open-source projects that will receive small grants to build Mojo and MAX libraries. The company said it received far more strong proposals than it could support.

The funded projects cover a broad slice of the emerging Mojo ecosystem, from reinforcement learning frameworks to JSON parsing and HTTP clients.

The six funded projects

noeira, created by Denis Laboureyras in early 2026, is an open-source reinforcement learning framework written entirely in Mojo. It supports more than 40 RL algorithms, GPU-accelerated deep RL, custom 2D and 3D physics engines, native arcade game engines and SDL3 rendering. The long-term goal is a fully end-to-end RL framework where environments, agents and training all run on the GPU in pure Mojo. In the near term, the project plans to publish benchmarks against established RL stacks including Gymnasium, Stable Baselines3 and PyTorch to quantify performance gains from a pure Mojo approach.

bajo, by Jean-Gabriel Simard, is a parallel physics simulator designed to run thousands of RL and robotics training environments simultaneously on a single GPU. It has already shipped a GPU sorting implementation using Morton encoding and BVH construction as its compute foundation. The six-month plan is to package that sorting layer, implement parallel collision detection and rigid-body dynamics, then scale to a multi-thousand-environment prototype and benchmark it against NVIDIA's Warp and Madrona.

Martin Vuyk Loperena's Fast Fourier Transform implementation is approaching state of the art performance on both CPU and GPU, competitive with FFTW and cuFFT. Started at a hackathon in mid-2025, the project has evolved continuously since. Near-term goals include adding half-sized transform support for hC2R and R2hC, and pushing for outright performance wins on the R2R, C2R and R2C transforms already supported. If the remaining gap closes, MAX's cuFFT dependency could be removed entirely.

EmberJson, by Brian Grenier, is the most widely used JSON parsing library in Mojo today. It leans heavily on Mojo's reflection capabilities and supports both structured and unstructured parsing. The roadmap points toward a general serialization framework in the spirit of Rust's Serde, plus continued work on Python interoperability. Early benchmarks show it outpacing Python's built-in JSON handling despite interop overhead.

floki, by Mikhail Tavarez, is a Mojo HTTP client built on libcurl bindings. It already supports many features familiar to users of Python's requests module. Development is now focused on ergonomic access to advanced libcurl configuration and simplifying the underlying bindings.

momanim, by Josiah Laivin, is a 2D animation framework for Mojo inspired by Manim. It aims to use CPU and GPU acceleration to dramatically reduce rendering times and targets an API close enough to Manim's that LLM-assisted Python-to-Mojo conversion is realistic. The project is in its infancy: basic image and video APIs and shape rendering are in place, with a 3D rendering API next on the roadmap.

What this means for the ecosystem

Modular framed the cohort as a snapshot of where the community is headed as Mojo and MAX develop, spanning foundational infrastructure and applications. For inference engineers and model builders, the notable threads are the RL and physics projects targeting GPU-native training loops, and the FFT work that could remove a cuFFT dependency from MAX. The JSON and HTTP libraries address everyday plumbing for serving and tooling workflows.

All six projects are open source, and Modular pointed to community meeting presentations for each contributor. The company said it is thankful to the community and looks forward to what comes next.