Trinity-Large-Thinking: Scaling Open-Source Frontier Agents

Arcee.ai has released Trinity-Large-Thinking, a frontier open reasoning model for complex long-horizon agents and multi-turn tool calling, under the Apache 2.0 license.

This model was designed to overcome the limitations of the previous Trinity-Large-Preview. By adopting a 'thinking' process before responding, it significantly strengthens multi-turn tool use, context consistency, instruction-following ability, and stability in long-horizon agentic loops.

Key performance and features are as follows:

• It ranked 2nd on the PinchBench agent benchmark, following Opus-4.6.
• When used via API, it costs about $0.90 per 1 million tokens, roughly 96% cheaper than competing models.
• It is provided as open weights so that developers and enterprises can directly inspect, post-train, host, and own the model.

Arcee.ai emphasizes that permissive open-weight models that can be inspected and owned are essential for US companies and developers to achieve true AI sovereignty, and through this release aims to provide a reliable and economical environment for building agents.