PulseAugur
LIVE 09:10:55
tool · [1 source] ·
32
tool

Zaya1-8B model beats GPT-5-High on math test without NVIDIA GPUs

A new language model named Zaya1-8B, featuring 760 million active parameters in a Mixture-of-Experts architecture, has demonstrated impressive performance on the HMMT '25 math competition. Notably, this model achieved its results without any training on NVIDIA GPUs, a significant departure from typical high-performance AI training. Zaya1-8B surpassed the performance of GPT-5-High on this specific math benchmark, scoring 89.6%. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Demonstrates novel training approaches can yield competitive results, potentially reducing reliance on expensive GPU infrastructure.

RANK_REASON The cluster reports on a new model's performance on a specific benchmark, which falls under research. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Towards AI →

COVERAGE [1]

  1. Towards AI TIER_1 · Chew Loong Nian - AI ENGINEER ·

    I Tested ZAYA1-8B — Trained on Zero NVIDIA GPUs, Its 760M Active Params Cheated GPT-5-High on Math

    <div class="medium-feed-item"><p class="medium-feed-snippet">A 760-million-active-parameter MoE that never touched a single NVIDIA H100 in training scored 89.6% on HMMT &#x2019;25 math &#x2014; 1.3 points higher&#x2026;</p><p class="medium-feed-link"><a href="https://pub.towardsa…