Arcee AI

What Arcee AI does

Arcee AI builds open-weight foundation models that run anywhere - on edge, on prem, or cloud. Their Trinity family standardizes capabilities across sizes and is released under Apache-2.0. Arcee AI is a U.S. model lab led by builders, focusing on efficiency, adaptability, and ownership of models. They optimize for performance per parameter and release weights under Apache-2.0, ensuring models are permanent and portable without lock-in or forced upgrades. Arcee invests in curated + synthetic data pipelines, reliability measured on tool use and structured outputs, and open-weight R&D to create a permanent, practical, efficient, trustworthy, and product-ready AI foundation.

Products

• Trinity Large ThinkingFrontier reasoning model for hard problems.
• Trinity MiniEveryday workhorse model with balanced performance and speed.
• Trinity NanoUltra-lightweight and fast enough to run on-device.
• TrinityAn open-weight model family that runs anywhere, providing consistent skills from on-device to cloud.
• OpenMDW-1.1Trinity is moving to this product.
• Hermes AgentAn agent that can be used with Trinity Large Thinking.

Key capabilities

• Open-weight foundation models
• Runs anywhere (edge, on-prem, cloud)
• Apache-2.0 license
• Consistent capabilities across model sizes
• Optimized for performance per parameter
• Permanent and portable models
• No lock-in or forced upgrades
• Curated + synthetic data pipelines
• Reliability measured on tool use and structured outputs
• Open-weight R&D
• API access for model deployment and calls

Use cases

• Frontier reasoning for hard problems
• Everyday tasks requiring balanced performance and speed
• On-device AI applications
• Long-running agents

AI approach

Arcee AI builds and releases open-weight foundation models, specifically the Trinity family, which are designed to run on various infrastructures including edge, on-premise, and cloud. They pre-train their own models, focusing on efficiency, adaptability, and performance per parameter, and release them under Apache-2.0. They also invest in curated and synthetic data pipelines and R&D for open-weight models.

Tech named: Trinity Large Thinking, Trinity Mini, Trinity Nano, Hermes Agent, OpenMDW-1.1, NVIDIA Blackwell, Apache-2.0

What it says sets it apart

• U.S. model lab
• Focus on open-weight models
• Apache-2.0 licensing for all models
• Models are permanent and portable
• No lock-in or forced upgrades
• Optimized for performance per parameter
• Commitment to open-weight R&D
• Consistent skills across different model sizes (on-device to cloud)

Funding rounds we track

From the AI funding tracker — rounds as reported by the linked publications.