Arcee AI

Arcee AI is a US-based open intelligence lab building open-weight language models with world-class performance per parameter.

Notable Models

Trinity Family — Sparse mixture-of-experts models for agents, tool use, and structured outputs (December 2025)

• Trinity Mini — 26B total, 3B active; cloud and on-prem production
• Trinity Nano — 6B total, 1B active; edge and embedded deployment

Arcee Foundation Models

• AFM-4.5B — 4.5B parameter instruction-tuned model; trained on 8T tokens

Legacy Models

• Arcee-Nova — 72B merge of Qwen2-72B-Instruct; general-purpose
• Virtuoso-Large — 72B general-purpose model
• Arcee-Spark — Small, fast model with 128K context
• Meraj-Mini — 7B Arabic/English bilingual model

Open Source Tools

• MergeKit — Model merging toolkit (LGPL v3)
• DistillKit — Knowledge distillation framework

Links

• Website
• Documentation
• Playground
• HuggingFace