ai-agents code-generation open-source-models reinforcement-learning

Ornith-1.0: self-improving open-source models for agentic coding

github.com

June 29, 2026

10 min read

🔥🔥🔥🔥🔥

60/100

Summary

Ornith-1.0 is an open-source self-improving model for agentic coding, available in configurations of 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE. It achieves state-of-the-art performance on coding benchmarks such as Terminal-Bench 2.1, SWE-Bench, NL2Repo, and OpenClaw by utilizing reinforcement learning for solution generation.

Key Takeaways

Ornith-1.0 is an open-source self-improving model for agentic coding, available in multiple configurations including 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE.
The model achieves state-of-the-art performance on coding benchmarks such as Terminal-Bench 2.1, SWE-Bench, NL2Repo, and OpenClaw, outperforming comparable open-source models.
Ornith-1.0 utilizes a self-improving training framework that employs reinforcement learning to optimize both solution rollouts and the scaffolds that drive those rollouts.
The model is MIT licensed, making it globally accessible and free from regional limitations.

Read original article

View story →

Coverage & discussions

Hacker News · github.com

Ornith-1.0: self-improving open-source models for agentic coding ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Mixed

Positives

Ornith-1.0 is the first Qwen fine-tune that has gained acceptance in the local LLM community, indicating a shift towards more reliable models for coding tasks.
Users have reported that Ornith-1.0 provides creative solutions to coding problems, showcasing its potential for practical applications in software development.
The model's accessibility for local hardware makes it a viable option for a broader audience, potentially democratizing AI tools for coding.

Concerns

There are concerns about the model's performance, particularly its tendency to hallucinate when used in chat without tools, raising questions about its reliability.
Critics argue that the title 'self-improving' is misleading, as the model's improvements stem from its training process rather than its operational capabilities.
Some users feel that the model is just a re-skinned version of existing models like Qwen or Gemma 4, suggesting a lack of innovation.