mak11boyz/Ornith-1.5-9B

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 27, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

Ornith-1.5-9B is a 9 billion parameter dense language model developed by mak11boyz, built upon Qwen3.5 and Gemma4, with a 32768 token context length. It features an advanced self-improvement loop for task generation, scaffold construction, and solution rollouts, making it highly optimized for coding, reasoning, and agentic tasks. This model is designed for efficient single-GPU deployment and is also available in a quantized mobile variant.

Loading preview...

Ornith-1.5-9B: Self-Improving Language Model

Ornith-1.5-9B is a 9 billion parameter model from mak11boyz, representing a significant advancement in self-improving foundation models. It builds upon its predecessor, Ornith-1.0 (which was based on Qwen3.5 and Gemma4), by expanding its self-improvement loop to jointly optimize task generation, scaffold construction, and solution rollouts. This allows the model to continuously generate new training tasks, discover effective problem-solving strategies, and refine its policy through reinforcement learning, moving beyond reliance on fixed human-curated tasks.

Key Capabilities & Differentiators

  • Advanced Self-Improvement: Utilizes a novel end-to-end self-improvement loop for continuous learning and optimization.
  • Strong Coding Performance: Achieves competitive results on coding benchmarks like Terminal-Bench 2.1 (46.2 Terminus-2, 47 Claude Code), SWE-bench Verified (70.6), SWE-bench Pro (47.5), and NL2Repo (32.4), often outperforming models in its size class and even larger models on specific tasks.
  • Enhanced Reasoning: Demonstrates improved reasoning capabilities, scoring 30.5 on HLE (with tools) and 86.4 on GPQA Diamond.
  • Agentic Proficiency: Excels in agentic tasks, with scores of 54.2 on MCP-Atlas, 41.2 on Toolathlon-Verified, and 66.5 on ClawEval.
  • Efficient Deployment: A 9B dense model designed for efficient single-GPU deployment, with a mobile-optimized variant for edge devices.
  • Long Context Support: Supports context windows up to 262,144 tokens, extendable to approximately 1M tokens using YaRN scaling.

Ideal Use Cases

  • Coding Assistants: Particularly strong for terminal-based coding agents, understanding large codebases, and automating development tasks.
  • Automated Reasoning: Suitable for applications requiring complex logical deduction and problem-solving.
  • Agentic Workflows: Excellent for integrating into agent frameworks that leverage tool calling and autonomous task execution.
  • Resource-Constrained Environments: Its efficient 9B parameter count and single-GPU deployment capability make it suitable for environments where larger models are impractical.