<a href="https://github.com/IQuestLab/IQuest-Coder-V1" target="_blank" rel="noopener nofollow">IQuest-Coder-V1</a> introduces a code-flow training approach that learns from commit history and code transformations, not just static snapshots.
The result? Models that better understand the dynamics of real projects, yielding stronger performance on benchmarks and practical coding tasks.
Available in sizes from 7B to 40B with native long-context support (up to 128K), Loop/Thinking variants for iterative reasoning, and deployment guides for Transformers and vLLM. Ideal for code generation, code understanding, and agentic developer workflows.
<img src="https://media.daily.dev/image/upload/s--udzL6un3--/f_auto/v1767248662/ugc/content_d85a9ace-93b1-4a7d-be70-ca96bfe27637?_a=BAMAK+ZW0" alt="G9joOhsaMAEhjD7">

Dokob

IQuest-Coder-V1 is a family of code models (7B to 40B parameters) trained using a code-flow approach that learns from commit history and code transformations rather than static code snapshots. The models support long contexts up to 128K tokens, include Loop/Thinking variants for iterative reasoning, and are designed for code generation, understanding, and agentic workflows. Deployment guides are available for Transformers and vLLM.

IQuest-Coder-V1 — High-Context, High-Performance Code Models (beats Claude 4.5 Sonnet)