Trinity-Large-Thinking
Arcee's open-weight (Apache 2.0) reasoning LLM tuned for long-horizon agents and tool use; ranks #2 on PinchBench and is cheap via API.
What it does
Trinity-Large-Thinking is an open-weight reasoning language model from Arcee AI, released under Apache 2.0 with weights on Hugging Face and access via Arcee's API. It 'thinks' before responding to improve multi-turn tool calling, instruction following and stability across long agent runs. Arcee reports it scores #2 on Kilo's PinchBench, just behind Opus 4.6, at roughly 96% lower output-token cost.
Core features
Open weights under Apache 2.0
Reasoning ('thinking') before responses
Strong multi-turn tool calling and long-context coherence
API access plus self-hosting
Competitive PinchBench score at low token cost
Best for
→Building agentic and tool-using applications
→Self-hosting a customizable frontier open model
→Cost-sensitive high-volume inference
Pricing
API
$0.90
Toolspool rankingby global site rank