Ornith 1.5: Free Coding Models & Ollama Setup

Ornith 1.5 is a free, MIT-licensed family of self-improving coding models from 9B to 397B. See sizes, benchmarks, Ollama setup and what hardware you need.

The Ornith team introduced Ornith 1.5 in August 2026 with three sizes, and added a fourth, a 122B mixture-of-experts model, on October 20, 2026. The idea behind it is self-improvement: the model proposes new coding tasks, writes the scaffolds to solve them and learns from its own attempts through reinforcement learning. The models are built on Qwen3.5 and Gemma 4 bases. If you searched for “ornith”, this is the current release. We checked the Ornith 1.5 post, the Ollama library page and the 122B model card on October 23, 2026.

Is Ornith 1.5 free?

Yes. Every Ornith 1.5 model is released under the MIT licence, so you can download the weights, run them on your own hardware and use them commercially. There’s no paid plan and no first-party API, so your only cost is the computer you run it on.

Ornith 1.5 Pricing (2026)

Ornith 1.5 costs $0 for every size. What changes between sizes is the hardware you need:

Model	Type	Ollama download	Good fit
Ornith 1.5 9B	Dense	6.6 GB	Laptops and 8 to 12 GB GPUs
Ornith 1.5 35B-A3B	MoE	23 GB	A 24 GB GPU or 32 GB Mac
Ornith 1.5 122B	MoE	73 GB (NVFP4) or 77 GB (Q4_K_M)	One 80 GB GPU or a 96 GB+ Mac
Ornith 1.5 397B	MoE	242 GB	Multi-GPU servers

If you don’t have the hardware, some third-party hosts list Ornith 1.5, but the Ornith team doesn’t run its own paid API, so check each host’s terms. For other free models you can run yourself, see our list of the best local AI models.

Ornith 1.5 benchmarks

These are the Ornith team’s own results on agentic coding tests:

Model	SWE-bench Verified	Terminal-Bench 2.1
Ornith 1.5 397B	86	86.1
Ornith 1.5 122B	84.5	81.2
Ornith 1.5 35B-A3B	79	67.8
Ornith 1.5 9B	70.6	46.2

The 9B model is the interesting one for most people. A 70.6 on SWE-bench Verified from a 6.6 GB download is close to what much bigger models scored a year ago. For comparison, Meta’s 30B Muse Glimmer is listed at 76.0.

How to run Ornith 1.5 with Ollama

Ollama is the quickest way to start. Install Ollama, then pull the size that fits your machine:

ollama run ornith-1.5:9b      # default, 6.6 GB
ollama run ornith-1.5:35b     # 23 GB
ollama run ornith-1.5:122b    # 73 to 77 GB
ollama run ornith-1.5:397b    # 242 GB

To use it as a coding agent, Ollama can launch it inside an agent app directly:

ollama launch claude --model ornith-1.5:9b
ollama launch opencode --model ornith-1.5:9b

Ollama lists the same launch command for Hermes Agent and OpenClaw. To call it from your own code, use Ollama’s local API:

curl http://localhost:11434/api/chat -d '{
  "model": "ornith-1.5:9b",
  "messages": [{"role": "user", "content": "Write a Python function that parses a CSV file"}]
}'

For servers, the 122B model card also gives a vLLM serve command, and it supports 262K context, extendable to about 1M tokens with YaRN. New to context windows? Our guide to what tokens are in AI explains them.

Limitations

• Self-reported benchmarks: the scores come from the Ornith team.
• No official API: you host it yourself or rely on third-party providers.
• Big downloads: the 122B and 397B models need server-class memory.
• Coding focus: it’s tuned for agentic coding, not general chat or writing.

Verdict

Ornith 1.5 is one of the best free coding models you can run at home. Start with the 9B in Ollama, move to 35B if you have a 24 GB GPU, and use the 122B on a single 80 GB card for near-frontier results. For a smaller general-purpose agent with image input, look at Muse Glimmer.

Ornith 1.5 gives you near-frontier coding scores in open, MIT-licensed models, from a 9B that runs on a laptop up to a 397B for servers.

Pull the 9B in Ollama to test it today, then size up to 35B or 122B if your hardware allows.