HortonArtificial Intelligence
Log InSign Up
Back to Models Overview

Horton T-Series

Our foundational model family, proudly developed in Germany, designed for privacy, efficiency, and performance.

Made in Germany AI

The Horton T-Series represents our commitment to building world-class AI from the heart of Europe. We believe in creating powerful, transparent, and auditable models that align with strong data privacy standards, providing a trustworthy alternative for developers and enterprises worldwide.

Horton T1 mini

The On-Device Powerhouse

A lightweight, powerful model designed for on-device AI features, offering speed and privacy without compromising capability.

T1 mini is designed to run locally on user devices, from mobile phones to powerful desktops. It will power features in upcoming products like HortonView, our on-device visual intelligence app, bringing powerful AI capabilities directly into users' hands without compromising their data.

Core Capabilities*
  • Efficient bilingual reasoning (English + German)
  • Lightweight architecture optimized for on-device use
  • Fast inference on Apple Silicon / edge devices
  • Privacy-focused fine-tuning pipeline (no external data sharing)
    • Technical Profile*

    Parameters*

    ≈ 2.1 billion

    Context Window*

    16k tokens (mobile) to 64k (desktop)

    Architecture*

    Decoder-only transformer

    Quantization*

    4-bit / 8-bit optimized

    Media Support*

    text

    Horton T1

    The Cloud-Based Foundational Model

    The Horton T-Series is our foundational model family, proudly developed in Germany. It includes the on-device T1 mini and the powerful, cloud-based T1, both architected for privacy and efficiency.

    As the larger model in the T-Series, T1 is a highly capable, cloud-based model designed to handle complex reasoning, analysis, and creative generation tasks at scale. It will be available through the Horton AI Studio, providing a powerful and cost-effective solution for enterprises.

    Core Capabilities*
  • Advanced multilingual reasoning (optimized for English and German, with extended European language support)
  • Deep contextual understanding across text, code, and visual input
  • High-efficiency distributed inference for large-scale cloud deployment
  • Reinforcement- and alignment-tuned for creative, analytical, and instructional tasks
  • Privacy-first architecture — zero data retention, fully isolated session memory
    • Technical Profile*

    Parameters*

    ≈ 38 billion

    Context Window*

    Up to 256k tokens (expandable via streaming memory)

    Architecture*

    Hybrid Mixture-of-Experts transformer with dynamic routing and adaptive attention layers

    Quantization & Deployment*

    Cloud inference in FP16 / BF16 precision. Edge compatibility through 8-bit adapters. Scalable multi-GPU sharding with latency-optimized attention kernels.

    Media Support*

    • text
    • code
    • image understanding

    Pricing

    Will be announced upon release