Agentic coding orientation
According to the model card, Ornith-1.0 is a self-improving family for agentic coding, with Ornith-1.0-9B focused on tool calling and coding capabilities.
Open Source Model Profile · ornith-ai
Ornith-1.0-9B
Ornith-1.0-9B is a dense agentic-coding model from ornith-ai. Its model card describes it as the lightweight 9B member of the Ornith-1.0 family for tool calling and coding work, with MIT licensing.
- Publisher
- ornith-ai
- Task
- text-generation
- Model type
- qwen3_5
- License
- mit
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Ornith-1.0-9B is published by ornith-ai as a text-generation model with captured architecture Qwen3_5ForConditionalGeneration. According to the model card, this is the most lightweight Ornith-1.0 member, a dense approximately 9B model designed for efficient single-GPU deployment and agentic coding. Card data records mit.
Recorded capabilities
Documented function calls
According to the model card, the model emits well-formed function calls parsed into the standard tool_calls field, with get_weather and run_shell style examples.
Single-GPU serving record
According to the model card, the dense approximately 9B model is about 19 GB in bf16 and serves on a single 80GB GPU, with recent Transformers, vLLM, and SGLang minimum versions documented.
MIT license record
Card data records mit, and the model card states MIT licensing with global access.
Use cases in the source record
- Agentic coding workflows using the card's documented tool-calling and OpenAI-compatible server setup.
- Local single-GPU serving experiments following the card's vLLM, SGLang, and Transformers recipes.
Limitations and unknowns
- Captured Safetensors metadata reports 1.47M parameters, which conflicts with the card's dense approximately 9B description; the discrepancy was not resolved in this record.
- According to the model card, benchmark figures such as 43.1 Terminal-Bench 2.1 Terminus-2, 69.4 SWE-bench Verified, and 42.9 SWE-bench Pro are publisher-reported results, not independently verified measurements.
- No context-window value was independently extracted, though serving examples use 262144 maximum length settings.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: ornith-ai/Ornith-1.0-9B
Captured: Unknown. Processed: 2026-09-07T19:34:54.495938+00:00.
Ornith-1.0-9B Aloha! 🌺 Today, we are releasing Ornith-1.0, a self-improving family of open-source models for agentic coding. Highlights: State-of-the-Art Coding Agents : Available in 9B-Dense, 31B-Dense, 35B-MoE, and 397B-MoE (post-trained on top of Gemma 4 and Qwen 3.5), achieving state-of-the-art performance among open-source models of comparable size on coding benchmarks such as Terminal-Bench 2.1, SWE-Bench, NL2Repo and OpenClaw. Self-Improving Training Framework : Ornith-1.0 employs RL to learn to generate not only solution rollouts, but also the scallfold that drive those rollouts. By jointly optimizing the scaffold and the r…
F001F002F003F004F005F006F007F010F011F012F014F015F016F017F018F019F022F024F025F026F028