Arcee Releases Trinity Large Thinking Open-Weight Model

Arcee, a 26-person U.S. startup, publicly released Trinity Large Thinking — a 400B-parameter open-weight reasoning model — and is pitching it as a Western, downloadable alternative to China-based open models. CEO Mark McQuade calls it the most capable open-weight model “ever released by a non-Chinese company.”

Trinity Large Thinking is framed as an agentic reasoning model intended for multi-turn, complex tasks. Arcee provides both downloadable weights for on-premises fine-tuning and a cloud-hosted API, preserving the typical open-source tradeoff: lower control compared with closed models but much greater deployment flexibility and auditability.

Arcee’s positioning emphasizes sovereignty and enterprise control — companies can train and host the model themselves rather than rely on remote-hosted Chinese or large closed-source models. The release arrives as some OpenClaw users lost dependable access to Anthropic’s Claude, creating immediate demand for alternative models; Arcee reports Trinity has become a top model among OpenClaw users. Arcee concedes its models do not yet outperform the highest-performing closed models from big labs, but it stresses the benefits of open weights and vendor independence.

For ML engineers and infra teams building on-prem agentic systems, Trinity lowers a practical barrier: an auditable, downloadable large model from a Western provider. That matters for regulated industries, governance-conscious deployments, and teams needing full-stack control over fine-tuning and inference. For researchers, another competitive open option pushes benchmarks and tooling compatibility.

third-party benchmarks comparing Trinity to top closed models, community tooling and adapters (OpenClaw adoption is an early signal), and whether Arcee sustains model quality and support as usage scales.