Language

ZAYA

Frontier reasoning language models, built for production speed.

ZAYA1 Reasoning

Built for long context and low latency.

Modalities

Large Language Model (LLM)

Architecture

8.3B-760M Mixture-of-Experts (MoE)

Features

State-of-the-Art Benchmarks

Competitive with Qwen3-4B and Gemma3-12B. Surpasses Granite-4-H-Tiny.

Fast Time-to-first-token

800M-class latency with 12B-class quality.

Trained Full Stack AMD

First model trained end-to-end on AMD’s hardware, software, and networking stack.

Model Components

Compressed Convolutional Attention (CCA)

Compresses inputs before attention without sacrificing quality.

ZAYA Router

Ensures true expert specialization while maintaining balance across hardware.

Previous Models

ZAYA1 Base