ManniX-ITA/JackOD-9B-Coder
ManniX-ITA/JackOD-9B-Coder is a 9 billion parameter language model, an omnimerge_v2 blend based on Qwen/Qwen3.5-9B, specifically engineered for multi-turn agentic coding. This model excels at staying on task across multiple turns and tool calls, demonstrating superior termination rates and autonomous planning capabilities compared to its base and source models. It is optimized to finish coding tasks efficiently, making it particularly effective for complex, iterative development workflows.
Loading preview...
JackOD-9B-Coder: Multi-Turn Agentic Coding Specialist
JackOD-9B-Coder is a 9 billion parameter model, an omnimerge_v2 blend derived from Qwen/Qwen3.5-9B and three other coding-focused models. Its core design targets multi-turn agentic coding, emphasizing the ability to persist through complex tasks involving multiple interactions and tool calls until completion. Unlike models optimized for broad knowledge benchmarks, JackOD-9B-Coder prioritizes task completion and efficient termination in coding scenarios.
Key Capabilities & Differentiators
- Superior Task Termination: Achieves a significantly lower rate of non-terminating generations (1 out of 55 hard LiveCodeBench problems) compared to its base (18x reduction) and heaviest source (25x reduction), indicating a strong ability to conclude tasks.
- Enhanced Autonomous Planning: Demonstrates a clear edge in autonomous planning, scoring 5.2/6 in
tool-eval-bench's 'Autonomous Planning' category, outperforming all its source models, including those with higher aggregate tool-calling scores. - Strong LiveCodeBench Performance: Achieves a LiveCodeBench v6 score of 0.7818, surpassing all its constituent models and the base, particularly on hard coding problems.
- Text-Only Model: Despite inheriting vision-related configurations, this model is text-only and does not possess a vision tower.
Ideal Use Cases
- Agentic Coding Workflows: Designed for scenarios requiring an AI agent to manage and complete coding tasks over multiple steps, tool interactions, and turns.
- Complex Code Generation & Refinement: Suitable for tasks where models need to iterate, self-correct, and persist until a functional solution is achieved.
- Reducing Runaway Generations: Beneficial for applications where controlled output length and reliable task termination are critical to avoid excessive resource consumption or irrelevant output.