Ornith is a new open-source model family for developers who want coding agents that can run through terminal tasks, tool calls, and repository-level workflows. This guide explains what Ornith is, how it works, and where it fits in an AI coding stack.
What Is Ornith?
Ornith, also called Ornith-1.0, is a family of open-source large language models from DeepReinforce.
It focuses on agentic coding. That means the model does more than answer coding questions. It can help a coding agent inspect files, call tools, reason through multi-step tasks, and work across a codebase.
DeepReinforce released Ornith in several model sizes, including 9B, 31B, 35B MoE, and 397B MoE. The family targets developers, AI agent builders, and teams testing open alternatives to closed coding models.
Overview
| Item | Details |
|---|---|
| Name | Ornith / Ornith-1.0 |
| Developer | DeepReinforce |
| Type | Open-source LLM family |
| Main use case | Agentic coding |
| Model sizes | 9B Dense, 31B Dense, 35B MoE, 397B MoE |
| License | MIT |
| Open-source access | Model checkpoints on Hugging Face |
| Local use | Supported through vLLM, SGLang, Transformers, GGUF builds, Ollama, and llama.cpp options |
| Best fit | Coding agents, terminal workflows, repository tasks, tool-using developer assistants |
Features
Agentic Coding Focus
Ornith targets coding agents rather than simple autocomplete. You can use it in workflows where the model needs to inspect a repo, plan edits, run tools, and return code changes.
Self-Scaffolding Training
DeepReinforce trained Ornith with a self-improving strategy. The model learns both the solution path and the scaffold that guides that path, instead of depending only on a fixed human-written harness.
In practice, this aims to help the model search through coding tasks with better task-specific structure.
Multiple Model Sizes
Ornith gives developers several deployment choices. The 9B model suits local and lighter use cases, while the 35B and 397B versions target stronger agentic performance with heavier compute needs.
This range helps you test the model before committing to larger infrastructure.
Tool Calling Support
The Ornith model card includes examples for OpenAI-compatible serving, tool calling, and agent framework integration. This makes it easier to plug Ornith into coding CLIs, MCP-style tools, or custom agent backends.
Long-Context Coding Workflows
The 35B model card shows serving examples with large context settings. Long context can help when a coding agent needs to read more files, compare modules, or keep task state during multi-step repository work.
Long context does not remove the need for testing. You should still run unit tests, lint checks, and human review before merging code.