News Hub

Modular 26.4 includes state-of-the-art MoE serving and Mojo 1.0 Beta 2.

MiniMax M3 open weights model is available on Modular Cloud.

Modular released Mojo 1.0 Beta, community Mojo libraries, and real-time patient conversations powered by MAX.

Part 3 of Modular's series on why LLM inference requires a new router architecture.

Three trends from MLSys 2026.

Article discusses why LLM inference requires a new kind of router.

Developer built a pure Mojo app and ten libraries using AI agents.

Hippocratic AI partners with Modular for real-time patient conversation inference.

Modular uses AI agents to translate code to Mojo.

Inkwell examines why inference platform choice matters as much as model selection.

Modular discusses why LLM inference requires a new router design in part one of a series.

Modular released version 26.3 featuring Mojo 1.0 Beta and MAX Video Gen.

Modular announced AMD AI DevDay participation, new offices, and community shipping updates.

Frontier Coding Agents built a video diffusion pipeline using Modular's MAX.

TileTensor provides safer and more efficient GPU kernel implementation.

Structured Mojo Kernels Part 4 addresses portability and future development.

Modular achieves fastest Gemma 4 performance on NVIDIA and AMD hardware.

Modular published part one on software pipelining techniques for GPU kernels.

Modular published part three on composition practices for structured Mojo kernels.

Modular 26.2 adds state-of-the-art image generation and upgraded AI coding with Mojo.