DeepReinforce has introduced Ornith-1.0, an open-source family of agentic coding models, including dense and mixture-of-experts variants. The 35B model is available through Hugging Face and Ollama with MIT licensing and deployment notes for local and server-based use.
DeepReinforce introduced Ornith-1.0 as an open-source family of language models designed for agentic coding tasks.
In its release page, DeepReinforce describes Ornith-1.0 as a set of “self-scaffolding” models for software-development agents. The family includes 9B Dense, 31B Dense, 35B MoE, and 397B MoE variants, according to DeepReinforce’s announcement.
The company positions the models for agentic coding workflows, including tasks that require planning, implementation, testing, and iterative refinement. DeepReinforce says the Ornith-1.0-35B model was evaluated on coding and agent benchmarks including Terminal-Bench 2.1, SWE-bench, NL2Repo, ClawEval, and SWE Atlas.
The release is notable because it targets a practical area of current model use: coding agents that can operate across repositories and terminal environments rather than only generating isolated snippets. However, the available public materials should be read as model-release documentation and benchmark reporting from the publisher, not independent proof of superiority over other coding models.
The official Hugging Face model card for deepreinforce-ai/Ornith-1.0-35B lists the model as MIT-licensed and optimized for coding agents. The model card provides usage notes for several deployment paths, including vLLM, SGLang, GGUF/Ollama, OpenHands, OpenCode, and OpenAI-compatible endpoints.
That makes the 35B release accessible to developers who want to test it in different environments, from local inference setups to agent frameworks. The Hugging Face documentation also indicates that DeepReinforce is treating the model as part of a developer-facing tooling ecosystem rather than only a standalone chat model.
Ollama’s model catalog also lists Ornith as a self-improving open-source model family for agentic coding. Ollama includes a runnable 35B tag and shows the command ollama run ornith:35b. The Ollama listing also cites a 256K context window for the 35B tag, which is relevant for repository-scale coding tasks where long context can matter.
Independent developer and writer Simon Willison covered the Ornith-1.0 release on his weblog, describing it as DeepReinforce’s first open-weights, MIT-licensed model release. Willison also reported running the 35B Q4_K_M GGUF version in LM Studio for agent-harness coding tasks.
Willison’s post is useful as an early outside account of hands-on testing, but it is not a broad benchmark study. It does, however, show that the 35B model is already being tried in local-model workflows by developers who evaluate coding models outside hosted APIs.
For teams or individual developers considering Ornith-1.0, the clearest verified facts are availability, licensing, and supported deployment notes. DeepReinforce has published the model family page; Hugging Face hosts the 35B model card with MIT licensing and integration guidance; and Ollama lists a runnable 35B version.
The more subjective question is where Ornith-1.0 fits among coding models such as Qwen-based systems or other open-weights alternatives. The supplied public sources do not establish a definitive ranking against those models. Developers interested in pairings such as a coding model plus a separate reviewer model would need to run their own task-specific evaluations for correctness, edge cases, architecture, style, security, and performance.
Ornith-1.0 therefore lands as a credible new open-source entrant in the agentic-coding model category, with enough public documentation for experimentation. Its real-world value will depend on how it performs in reproducible development tasks, repository-specific tests, and coding-agent harnesses used by practitioners.
DeepReinforce introduced Ornith 1.0 as an open source family of language models designed for agentic coding tasks.
A coding focused model family In its release page, DeepReinforce describes Ornith 1.0 as a set of “self scaffolding” models for software development agents.
The family includes 9B Dense, 31B Dense, 35B MoE, and 397B MoE variants, according to DeepReinforce’s announcement.
Continue reading
