Function Calling With Schema Based Expert Routing

via “function-calling with schema-based expert routing”

Trinity Mini is a 26B-parameter (3B active) sparse mixture-of-experts language model featuring 128 experts with 8 active per token. Engineered for efficient reasoning over long contexts (131k) with robust function...

Unique: Leverages sparse MoE architecture where certain experts can specialize in tool-use reasoning, potentially improving function-calling accuracy through expert specialization — most competing models use uniform dense layers for all reasoning types

vs others: Maintains function-calling accuracy comparable to GPT-4 and Claude while operating at 3B effective parameters, reducing inference costs by 5-10x for tool-using agent applications