IQuest Releases IQuest-Q1: A 320B MoE Model for Agentic Coding
IQuest has released IQuest-Q1, an open-weights Mixture-of-Experts (MoE) language model designed for agentic coding, multi-step reasoning, and tool use. The model features approximately 320 billion total parameters, with 15 billion parameters activated per token during inference. By leveraging an MoE architecture, IQuest-Q1 delivers the reasoning performance of a massive model while maintaining the computational efficiency of a much smaller 15B model. This gives open-source AI developers a capable base model for complex software engineering agents and autonomous workflows. The model activates only around 4.7% of its total parameter count per token, optimizing generation speed and hardware resource efficiency. It is specialized for multi-step tool execution, complex logical reasoning, and autonomous code synthesis.
## BACKGROUND
A Mixture-of-Experts (MoE) architecture scales large language models efficiently by routing each input token to a specialized subset of expert sub-networks rather than processing through all parameters. Agentic coding is an approach to software development where autonomous AI agents independently plan, write, execute, and test code using external tools with minimal human supervision.