Arcee AI

Trinity Large Thinking

Arcee AI

Trinity Large Thinking is Arcee AI’s reasoning-focused post-training of Trinity Large Base. The sparse mixture-of-experts model has 398B total parameters and approximately 13B active per token, with documented context extension to 512K. It generates explicit thinking traces and is trained for long-running agents, multi-turn tool calls, and coherent instruction following. Thinking tokens must remain in conversation history for supported agent behavior. API access and downloadable weights are available. The current model card specifies OpenMDW-1.1, despite Apache 2.0 wording in the original announcement; this record follows the current weight license.

Trinity Large Preview

Arcee AI

Trinity Large Preview is a distinct size tier in Arcee AI’s Trinity family of sparse language models trained from scratch. It combines a larger total weight capacity with relatively few active parameters per token. Large Preview is lightly post-trained for chat and long-context comprehension, with separate base and reasoning checkpoints. Downloadable weights and deployment examples support private inference and customization. The current model card specifies OpenMDW-1.1; deployment rights should follow that release license.

Trinity Mini

Arcee AI

Trinity Mini is a distinct size tier in Arcee AI’s Trinity family of sparse language models trained from scratch. It combines a larger total weight capacity with relatively few active parameters per token. Mini targets efficient reasoning and tool-driven applications. Downloadable weights and deployment examples support private inference and customization. The current model card specifies OpenMDW-1.1; deployment rights should follow that release license.

Trinity Nano Preview

Arcee AI

Trinity Nano Preview is a distinct size tier in Arcee AI’s Trinity family of sparse language models trained from scratch. It combines a larger total weight capacity with relatively few active parameters per token. Nano is an experimental chat checkpoint whose aggressive sparsity can produce unstable behavior. Downloadable weights and deployment examples support private inference and customization. The current model card specifies OpenMDW-1.1; deployment rights should follow that release license.