Featured Releases

Instella-Math

First Math LLM with Long CoT RL on AMD GPUs

Learn more →

Instella-T2I

Pushing the Limits of 1D Discrete Latent Space Image Generation

Learn more →

Instella-Long

A Fully Open Language Model with Long-Context

Learn more →

AMD GenAI advances the frontier of AI through open innovation and collaboration. We build fully open-source foundation models optimized for AMD GPUs such as MI250 and MI300, and release everything -- code, models, and training recipes -- to the public.

Our mission is to empower AI researchers and engineers to train and develop cutting-edge models efficiently on AMD systems.

AMD GenAI is part of AMD. To learn more about AMD, visit AMD.

Recent Blogs
Instella-Math
August 9, 2025
Introducing Instella-Math — AMD’s first fully open reasoning language model, trained end-to-end with long chain-of-thought reinforcement learning on 32 MI300X GPUs. Optimized for complex reasoning and math, with all code, weights, and datasets released for the community.
GEAK: Triton Kernel AI Agent
August 1, 2025
Introducing GEAK, AMD’s AI-driven Triton kernel generation framework and evaluation benchmarks. GEAK leverages frontier LLMs with inference-time scaling to automatically produce efficient, accurate GPU kernels for AMD Instinct™ GPUs, achieving up to 63% correctness and 2.59× speedups over reference implementations.
Instella-T2I
July 15, 2025
Introducing 1D binary image latents for text-to-image generation, achieving up to 32× token reduction and competitive 1024×1024 performance with faster, scalable training and inference.
Read All Blogs
Recent Publications
SAND-Math: Using LLMs to Generate Novel, Difficult and Useful Mathematics Questions and Answers
Publication
2025
SAND-Math: Using LLMs to Generate Novel, Difficult and Useful Mathematics Questions and Answers
SAND-Math, a scalable pipeline that generates and enhances challenging math problems, enabling LLMs to achieve state-of-the-art results on difficult mathematical reasoning benchmarks like AIME25.
Read more
TTT-Bench: A Benchmark for Evaluating Reasoning Ability with Simple and Novel Tic-Tac-Toe-style Games
Publication
2025
TTT-Bench: A Benchmark for Evaluating Reasoning Ability with Simple and Novel Tic-Tac-Toe-style Games
TTT-Bench uncovers the hidden blind spots of today’s smartest AI, challenging them with simple, human-intuitive games — and showing that even the best models often fail where humans excel effortlessly.
Read more
TaDA: Training-free recipe for Decoding with Adaptive KV Cache Compression and Mean-centering
Publication
ACL 2025 Industry Track (Oral)
TaDA: Training-free recipe for Decoding with Adaptive KV Cache Compression and Mean-centering
TaDA slashes KV cache memory usage by over 70% without sacrificing accuracy — enabling longer, smarter, and more scalable LLM inference with zero retraining.
Read more
Read All Publications