TAAFT
Free mode
100% free
Freemium
Free Trial
Deals

ZAYA1-8B

ZAYA1-8B is a post-trained reasoning version of Zyphra’s small MoE language model, designed to deliver unusually strong math, coding, and detailed reasoning for its size. The model card says it has 8.4B total parameters with 760M active parameters, is trained end-to-end by Zyphra, and uses architectural plus training innovations to push performance far above similarly sized models. Zyphra positions it as especially effective for long-form reasoning and test-time-compute setups, while also being small enough for local on-device LLM applications.
New Multimodal Gen 3
Released: May 6, 2026

Overview

ZAYA1-8B is Zyphra’s small mixture-of-experts reasoning model, built for long-form reasoning, mathematics, and coding. It has 8.4B total parameters with 760M active parameters, is post-trained for reasoning, and is positioned as highly intelligence-efficient enough for both strong test-time-compute use and local on-device deployment.

About Zyphra AI

Superintelligence that Empowers.
Zyphra is a full stack open source superintelligence company based in San Francisco, California.
Our mission is to build human-aligned AI that helps individuals and organizations reach their fullest potential.

Industry: Artificial Intelligence
Company Size: 67
Location: Palo Alto, CA, US
Website: zyphra.com
View Company Profile

Tools using ZAYA1-8B

No tools found for this model yet.

Last updated: May 7, 2026
0 AIs selected
Clear selection
#
Name
Task