iBossonline/RegesCore-2.1-9B

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 1, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

iBossonline/RegesCore-2.1-9B is the Ornith-1.0-9B model, a 9 billion parameter dense language model developed by DeepReinforce. It is specifically designed for agentic coding tasks, excelling on benchmarks like Terminal-Bench 2.1, SWE-Bench, NL2Repo, and OpenClaw. This model utilizes a self-improving training framework based on RL to optimize solution generation and search trajectories, making it highly effective for automated code development and problem-solving.

Loading preview...

Ornith-1.0-9B: Agentic Coding Model

Ornith-1.0-9B, developed by DeepReinforce, is a 9 billion parameter model optimized for agentic coding. It is the most lightweight member of the Ornith family, designed for efficient single-GPU deployment. The model employs a unique self-improving training framework that uses Reinforcement Learning (RL) to generate both solution rollouts and the scaffolds driving them, leading to enhanced search trajectories and higher-quality solutions.

Key Capabilities

  • State-of-the-Art Agentic Coding: Achieves leading performance among open-source models of comparable size on coding benchmarks such as Terminal-Bench 2.1 (43.1 on Terminus-2, 40.6 on Claude Code), SWE-Bench (69.4 Verified, 42.9 Pro, 52 Multilingual), NL2Repo (27.2), and OpenClaw (63.1 Avg).
  • Self-Improving Framework: Learns to discover better search trajectories and generate superior solutions by jointly optimizing the scaffold and the resulting solution.
  • Reasoning Model: By default, the assistant turn opens with a <think>...</think> block, providing a chain-of-thought that can be parsed separately. It also supports tool-call parsing for OpenAI-style tool_calls.
  • Efficient Deployment: Designed for single-GPU deployment, requiring approximately 19 GB in bf16.
  • Open-Source License: MIT licensed, ensuring global accessibility and freedom from regional limitations.

Good for

  • Automated Code Generation: Excels at generating code and solving complex programming problems through agentic approaches.
  • Tool-Calling and Agent Frameworks: Works seamlessly with standard agent frameworks and OpenAI-compatible endpoints, supporting tool use for interactive problem-solving.
  • Terminal-Based Coding Agents: Optimized for use in coding CLIs to understand large codebases, automate tasks, and accelerate development workflows.
  • Research and Development: Its self-improving training framework offers a valuable resource for researchers exploring advanced RL techniques in language models.

Finetunes
1 models