DeepReinforce Launches Ornith-1.0, Tailored for Coding Agents

Overview

• DeepReinforce introduced Ornith-1.0 on June 25, aimed at enhancing AI capabilities for coding tasks in real-world environments.
• The 9 billion parameter model scored 69.4 on SWE-bench Verified, outperforming Google's Gemma 4-31B, which scored 52.0.
• However, Ornith-1.0 is specifically designed for coding and may not perform well on other tasks.

DeepReinforce, a company recognized for its innovative projects like CUDA-L1, has launched Ornith-1.0, a suite of open-source models available on Hugging Face. The new models come in various sizes, including 9 billion, 31 billion, and a top-tier 397 billion mixture-of-experts model, all released under the MIT license, making them accessible without regional restrictions.

Ornith-1.0 represents a specialized collection of coding models optimized for agentic tasks, allowing AI to autonomously manage coding-related responsibilities such as reading files, running tests, and correcting code. Unlike traditional AI, these coding agents operate independently, minimizing the requirement for human oversight.

Introducing Ornith-1.0: a suite of open-source models tailored for agentic coding.

Check out our latest models achieving cutting-edge performance! pic.twitter.com/7g1rmacLps

— Ornith (@ornith_) June 25, 2026

Ornith’s architecture allows it to adaptively create strategies rather than following fixed instructions, honing its approach through a two-step reinforcement learning process. This involves proposing improved methods based on task analysis and iterating solutions while using feedback to refine subsequent strategies.

Performance-wise, the 397 billion parameter model scored 82.4 on the SWE-bench Verified assessment, surpassing other high-profile models like Claude Opus 4.7 and DeepSeek-V4-Pro. Its performance on the more challenging SWE-bench Pro was also notable, scoring 62.2. Although the smaller 9 billion parameter model delivered impressive results, it is tailored for coding tasks specifically and might not be suitable for broader applications such as document summarization or academic writing.

The launch of Ornith-1.0 aligns with the increasing demand for AI that can handle complex programming workflows autonomously, which could potentially revolutionize developer productivity.