IQuest-Coder-V1-40B
Open-source code intelligence model — 81.4% on SWE-Bench Verified
An open-source code intelligence model scoring 81.4% on SWE-Bench Verified.
Open foundation models — the base layers everything builds on.
Key features
- Code LLM family from 7B to 40B with 128K native context
- Code-flow multi-stage training; 76.2% on SWE-Bench Verified
- Thinking and Instruct variants; recurrent Loop architecture
- CLI agent integration with Claude Code and OpenCode
- Preliminary HTML and SVG generation support
Good for
- Agentic software engineering and code generation
- CLI-based coding agent workflows
Stack & tags
code-llmcoding-agentopen-weightsswe-benchlong-contextthinkingcli