Paper page - ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search

Source: https://huggingface.co/papers/2609.13356 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

ZGCM-1 is a 7B open foundation model that combines internal reasoning with external tool use, trained via efficient architecture-system co-design, progressive long-context scaling, and autonomous agent workflows to achieve strong reasoning and efficiency.

In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency. ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can overcome parametric capacity limits by coupling deliberate internal thinking with active external tool use. To support this paradigm across a 256K context, we develop an end-to-end, high-efficiency open training recipe: Architecture & System Co-design: interleaved gated sliding-window andfull attention, and a stableFP8 Muon optimizer; Progressive Curriculum & MDPMid-Training: context scaling across 16K, 64K, and 256K, and the reformulation of interaction traces intoMarkov Decision Processes. Furthermore, we establish an AI-native R&D workflow whereagent swarmsautonomously manage cluster operations, data curation, and rapid diagnostic evaluation. Extensive evaluations show that ZGCM-1-7B is competitive across 7B model family on general benchmarks. On several challenging mathematical reasoning and agentic search suites, it remains competitive with frontier models orders of magnitude larger, such as Qwen3-235B-A22B and GLM-5.1. We also show that ourpre-trainingdesign offers a ~4.2x efficiency improvement in 16Kpre-trainingtime-to-loss. Across the full development lifecycle, we distill eight actionable empirical findings-spanning architectural scaling, SFT quality pruning, long-context generalization, andagentic co-trainingdynamics. To facilitate community research, we open-source model weights from thepre-training,mid-training, andpost-trainingstages, intermediate checkpoints, training code, per-stage data and data recipes, and W&B logs.

View arXiv pageView PDFProject pageGitHub75Add to collection

Get this paper in your agent:

hf papers read 2609\.13356

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.13356 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.13356 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.13356 in a Space README.md to link it from this page.

Collections including this paper1