IQuest Research Releases IQuest-Q1 for Coding and Agent Tasks

IQuest Research released IQuest-Q1, an open weight model for coding and agent tasks, the lab said in a press release. Its weights and technical materials are available through GitHub, Hugging Face and the IQuest website.

IQuest-Q1 uses a decoder only Transformer with a sparse mixture of experts architecture. It has about 320 billion total parameters, with about 15 billion active for each token. Training covered code, reasoning, tool use, long context processing and multistep execution.

The lab demonstrated the model generating interactive applications from natural language prompts, diagnosing an error in a reinforcement learning run and completing tasks across workplace tools. Its published evaluations cover repository code generation, cybersecurity, terminal operation, long horizon coding and office workflows.