Saved in:
Bibliographic Details
Main Authors: Washbourne, Robert, Iyer, Rishi, Figliolia, Tomas, Zheng, Henry, Lorig-Roach, Ryan, Yang, Sungyeon, Yuvraj, Pritish, Anthony, Quentin, Tokpanov, Yury, Yang, Xiao, Nanduru, Ganesh, Ebert, Stephen, Medepalli, Praneeth, Szot, Skyler, Rajagopal, Srivatsan, Ong, Alex, Mehta, Bhavana, Millidge, Beren
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.05365
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918486836510720
author Washbourne, Robert
Iyer, Rishi
Figliolia, Tomas
Zheng, Henry
Lorig-Roach, Ryan
Yang, Sungyeon
Yuvraj, Pritish
Anthony, Quentin
Tokpanov, Yury
Yang, Xiao
Nanduru, Ganesh
Ebert, Stephen
Medepalli, Praneeth
Szot, Skyler
Rajagopal, Srivatsan
Ong, Alex
Mehta, Bhavana
Millidge, Beren
author_facet Washbourne, Robert
Iyer, Rishi
Figliolia, Tomas
Zheng, Henry
Lorig-Roach, Ryan
Yang, Sungyeon
Yuvraj, Pritish
Anthony, Quentin
Tokpanov, Yury
Yang, Xiao
Nanduru, Ganesh
Ebert, Stephen
Medepalli, Praneeth
Szot, Skyler
Rajagopal, Srivatsan
Ong, Alex
Mehta, Bhavana
Millidge, Beren
contents We present ZAYA1-8B, a reasoning-focused mixture-of-experts (MoE) model with 700M active and 8B total parameters, built on Zyphra's MoE++ architecture. ZAYA1-8B's core pretraining, midtraining, and supervised fine-tuning (SFT) were performed on a full-stack AMD compute, networking, and software platform. With under 1B active parameters, ZAYA1-8B matches or exceeds DeepSeek-R1-0528 on several challenging mathematics and coding benchmarks, and remains competitive with substantially larger open-weight reasoning models. ZAYA1-8B was trained from scratch for reasoning, with reasoning data included from pretraining onward using an answer-preserving trimming scheme. Post-training uses a four-stage RL cascade: reasoning warmup on math and puzzles; a 400-task RLVE-Gym curriculum; math and code RL with test-time compute traces and synthetic code environments built from competitive-programming references; and behavioral RL for chat and instruction following. We also introduce Markovian RSA, a test-time compute method that recursively aggregates parallel reasoning traces while carrying forward only bounded-length reasoning tails between rounds. In TTC evaluation, Markovian RSA raises ZAYA1-8B to 91.9\% on AIME'25 and 89.6\% on HMMT'25 while carrying forward only a 4K-token tail, narrowing the gap to much larger reasoning models including Gemini-2.5 Pro, DeepSeek-V3.2, and GPT-5-High.
format Preprint
id arxiv_https___arxiv_org_abs_2605_05365
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ZAYA1-8B Technical Report
Washbourne, Robert
Iyer, Rishi
Figliolia, Tomas
Zheng, Henry
Lorig-Roach, Ryan
Yang, Sungyeon
Yuvraj, Pritish
Anthony, Quentin
Tokpanov, Yury
Yang, Xiao
Nanduru, Ganesh
Ebert, Stephen
Medepalli, Praneeth
Szot, Skyler
Rajagopal, Srivatsan
Ong, Alex
Mehta, Bhavana
Millidge, Beren
Artificial Intelligence
Computation and Language
We present ZAYA1-8B, a reasoning-focused mixture-of-experts (MoE) model with 700M active and 8B total parameters, built on Zyphra's MoE++ architecture. ZAYA1-8B's core pretraining, midtraining, and supervised fine-tuning (SFT) were performed on a full-stack AMD compute, networking, and software platform. With under 1B active parameters, ZAYA1-8B matches or exceeds DeepSeek-R1-0528 on several challenging mathematics and coding benchmarks, and remains competitive with substantially larger open-weight reasoning models. ZAYA1-8B was trained from scratch for reasoning, with reasoning data included from pretraining onward using an answer-preserving trimming scheme. Post-training uses a four-stage RL cascade: reasoning warmup on math and puzzles; a 400-task RLVE-Gym curriculum; math and code RL with test-time compute traces and synthetic code environments built from competitive-programming references; and behavioral RL for chat and instruction following. We also introduce Markovian RSA, a test-time compute method that recursively aggregates parallel reasoning traces while carrying forward only bounded-length reasoning tails between rounds. In TTC evaluation, Markovian RSA raises ZAYA1-8B to 91.9\% on AIME'25 and 89.6\% on HMMT'25 while carrying forward only a 4K-token tail, narrowing the gap to much larger reasoning models including Gemini-2.5 Pro, DeepSeek-V3.2, and GPT-5-High.
title ZAYA1-8B Technical Report
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2605.05365