OrionLLM/OxCoder-9B

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 7, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Featherless Exclusive Cold

OxCoder-9B is a 9 billion parameter coding model developed by OrionLLM, designed for long-horizon agentic coding and reasoning tasks. It features a 262,144 token native context window, enabling it to handle complex codebases and iterative development. The model excels in agentic terminal coding, error recovery, and front-end reasoning, outperforming larger models in its class on various coding and agentic benchmarks. It is particularly optimized for real-world software engineering scenarios and multi-tool orchestration.

Loading preview...

OxCoder-9B: A Compact Agentic Coding Model

OxCoder-9B, developed by OrionLLM, is a 9 billion parameter model specifically engineered for advanced agentic coding and reasoning. Despite its compact size, it demonstrates strong performance in agentic terminal and coding tasks, often rivaling much larger models.

Key Capabilities and Features

  • Agentic Performance: Trained on frontier agent traces from systems like Fable-5.1 and GLM-5.3, it excels in real-world software engineering scenarios.
  • Extended Context Window: Features a substantial 262,144 native token context window, allowing it to manage extensive, multi-file codebases and complex, long-horizon tasks.
  • Robust Error Recovery: Incorporates read-before-write patterns, responds to Language Server Protocol (LSP) diagnostics, and applies minimal edit diffs for precise iterative coding.
  • Front-end Reasoning: Exhibits strong understanding of UI logic, component architecture, and web-native patterns, a rare capability for models under 10 billion parameters.

Performance Highlights

OxCoder-9B consistently leads its size class across various benchmarks:

  • Agentic Terminal Coding: Achieves 49.6 on Terminal-Bench 2.1 (Terminus-2) and 50.8 on Terminal-Bench 2.1 (Claude Code).
  • Agentic Coding: Scores 73.5 on SWE-bench Verified and 49.1 on SWE-bench Pro.
  • Repo-level Code Generation: Attains 36.2 on NL2Repo.
  • Agentic Tasks: Demonstrates strong results in multi-tool orchestration (56.7 on MCP-Atlas), web browsing (57.4 on BrowseComp), and real-user agentic coding (67.8 on ClawEval).

Ideal Use Cases

OxCoder-9B is particularly well-suited for:

  • Automated software development and bug fixing.
  • Agentic systems requiring long-context code understanding and generation.
  • Front-end development tasks and UI logic implementation.
  • Iterative coding environments where precise error recovery and minimal diffs are crucial.

Finetunes
1 models