Arcee AI

Arcee AI is a US-based open intelligence lab that builds open-weight foundation models designed to run anywhere—on edge devices, on-premises infrastructure, or in the cloud. The company focuses on creating efficient, adaptable AI models released under Apache-2.0 license, with their Trinity family of models standardizing capabilities across different sizes. Arcee emphasizes performance per parameter rather than raw model size, enabling frontier-level performance without frontier pricing.

• Open-Weight Models - All models are released under Apache-2.0 license, providing full transparency and control with no vendor lock-in or forced upgrades
• Trinity Model Family - A comprehensive family of foundation models with consistent skills from on-device to cloud deployment, including Trinity Mini, Trinity Medium, and Trinity Large variants
• Online Reinforcement Learning - Models feature continuous learning capabilities, allowing deployment once with ongoing improvements through rapid iteration
• Cost-Efficient Architecture - Optimized architectures keep inference costs low while maintaining frontier-level performance benchmarks
• Playground Access - Users can interact with models directly through the chat playground at chat.arcee.ai to test capabilities before integration
• API Access - Developers can integrate Arcee models via API for production applications with comprehensive documentation available
• Open Source Catalog - Browse and access the full catalog of open-source models through the website or Hugging Face
• Enterprise Solutions - Custom enterprise deployments available for organizations requiring dedicated support and infrastructure
• DistillKit - Open-source tool for knowledge distillation, enabling efficient model compression and transfer learning
• Multi-Platform Deployment - Models designed for flexible deployment across edge devices, on-premises servers, and cloud environments

To get started, visit the Arcee AI playground to experiment with models, or access the API through the documentation portal. The open-source catalog on Hugging Face provides direct access to model weights for self-hosted deployments.