ManniX-ITA/JackOD-9B-Coder

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 11, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

JackOD-9B-Coder by ManniX-ITA is a 9 billion parameter model based on the Qwen3.5-9B architecture, specifically designed for multi-turn agentic coding tasks. This omnimerge_v2 model excels at completing complex coding problems and terminating appropriately, significantly reducing runaway generations compared to its base models. It demonstrates strong performance on LiveCodeBench v6 hard problems and shows a clear edge in autonomous planning for tool-use scenarios. Its primary strength lies in its ability to stay on task across multiple turns and tool calls until completion.

Loading preview...

JackOD-9B-Coder: A Specialized Agentic Coder

ManniX-ITA's JackOD-9B-Coder is a 9 billion parameter model built on the Qwen3.5-9B architecture, created through a four-way omnimerge_v2 process. Its core design goal is multi-turn agentic coding, focusing on task completion across multiple interactions and tool calls, rather than general knowledge benchmarks.

Key Capabilities and Differentiators

  • Optimized for Agentic Coding: Specifically engineered to finish coding tasks and stop appropriately, addressing a common issue of runaway generations in other models.
  • Superior Termination Rate: Achieves a significantly lower rate of non-terminating generations (1 out of 55 on LiveCodeBench v6 hard problems) compared to its base (18/55) and heaviest source (25/55).
  • Strong LiveCodeBench Performance: Outperforms all its source models and the base on the challenging LiveCodeBench v6 (55 hard) benchmark, scoring 0.7818.
  • Enhanced Autonomous Planning: Demonstrates a clear advantage in autonomous planning (5.2/6) in tool-calling benchmarks, a critical aspect for complex agentic workflows.
  • Text-Only Model: Despite inheriting vision-related configurations, this model is strictly text-only with no vision tower or capabilities.
  • Robust Tool Calling: While aggregate tool-calling scores are competitive, its strength lies in planning, ensuring effective tool utilization over multi-step processes.

When to Use This Model

JackOD-9B-Coder is ideal for applications requiring a coding agent that can reliably execute multi-turn tasks, interact with tools, and manage complex problem-solving workflows without prematurely stopping or generating excessively long, unhelpful outputs. It is particularly suited for scenarios where task completion and efficient resource usage (due to better termination) are paramount.