Modular

The company supports agent loops with tool calling, structured output, and hardware-portable serving on NVIDIA and AMD. It serves custom agent architectures built in MAX + Mojo in its cloud or a customer's cloud.

Run TTS, speech-to-text, and voice models with performance on NVIDIA and AMD. Ranked #1 on Artificial Analysis for TTS with Inworld. Custom voice model support with Mojo kernels.

The page says FLUX.2 images can be generated in under one second at $0.009 per image, 7× faster than PyTorch torch.compile, on NVIDIA or AMD GPUs, with deployment in the customer's cloud or the company's.

Serve Kimi, DeepSeek, Qwen, and custom code models on NVIDIA and AMD. Compiler-native speculative decoding, structured output at graph speed, and 10x lighter runtime for scaling millions of completions.

Thoughtful perspectives on high-performance AI, modern infrastructure, and the tools shaping the next generation of computing.

AMD and TensorWave invited Modular to collaborate on a presentation for AMD’s Media Tech Day and asked for a demo of MAX on AMD Instinct MI355. Modular did not have access to an MI355.

Since August 2023, the Mojo repository has included a small benchmark example titled nbody.mojo. This code is based on an example from The Computer Language Benchmarks Game, a site that benchmarks implementations of different algorithms in popular programming languages.

New licensing terms for MAX and Mojo that allows for unlimited non-commercial usage

The company offers AI software that users can deploy and manage themselves for free, with enterprise scaling plans.

Mojo is built on the latest MLIR compiler technology, an evolution of LLVM which Rust lowers to. This enables programmers to write code optimized for different CPU architectures, and use the same ergonomic programming model to compile and run native GPU kernels

Modular provides a platform for developing and deploying generative AI across CPUs, GPUs, and other types of hardware.

The unified AI inference stack - from custom GPU kernels to production cloud serving on NVIDIA and AMD. 2x performance. Top open models. Open source stack.

The privacy policy governs privacy rights across Modular-branded websites, products, and services.

A high-performance inference engine to build, optimize, and deploy AI apps fast. Run open models, scale across GPUs, and tap into CPU+GPU performance with Mojo.

Qualcomm CEO Cristiano Amon says AI agents could trigger a smartphone upgrade supercycle by using apps, understanding personal context, and acting for users. He says AI companies are asking Qualcomm to support models with at least 100 billion parameters on phones by 2028, while he sees smart glasses as the leading AI wearable. Qualcomm is also pursuing energy-efficient data-center chips, an open software platform through its Modular acquisition, and 6G applications such as device-based sensing.

Oct 2026 · SourcesNews article