Ornith 1.0

Ornith's first agentic-coding family, post-trained from Qwen3.5 and Gemma 4 in June 2026 under MIT. Replaced by Ornith 1.5.

In short. Ornith 1.0 is the first release of DeepReinforce's coding-agent models, in 9B, 35B and 397B sizes under the MIT license. The company has since replaced it with Ornith 1.5, which has the same sizes and higher scores, but the 1.0 files are still downloaded a lot.

Summary

DeepReinforce released Ornith 1.0 in June 2026 as a self-improving family for agentic coding. The announcement says it was developed on top of Qwen3.5 and Gemma 4 with continued pretraining, mid-training and post-training, and lists 9B and 31B dense models and 35B and 397B mixture-of-experts models. The Hugging Face organization holds the 9B, 35B and 397B checkpoints, with FP8 builds of the two larger ones and GGUF builds of the 9B and 35B, and no 31B repository. The config files match Qwen3.5-9B, Qwen3.5-35B-A3B and Qwen3.5-397B-A17B, and the context is 256K tokens.

The idea was to let reinforcement learning optimize the scaffold that drives a solution along with the solution itself, so the model learns to write the orchestration around its own work. Ornith reports 77.5 on Terminal-Bench 2.1 and 82.4 on SWE-bench Verified for the 397B model, against 70.3 and 80.8 for Claude Opus 4.7, 66.0 and 80.5 for MiniMax M3, and 67.9 and 80.6 for DeepSeek-V4-Pro. The 35B model scores 64.2 on Terminal-Bench 2.1 and 75.6 on SWE-bench Verified, and the 9B model 43.1 and 69.4. These numbers are Ornith's own.

Ornith 1.5 raises the scores at every size, so new work should start there. The 1.0 models stay popular because the small ones are easy to run. In the 30 days to 11 November 2026, Hugging Face showed about 2.4 million downloads of the 9B GGUF build and 1.5 million of the 35B GGUF build. The license is MIT and, per the project, has no regional limits.

Good for

  • Reproducing results or keeping an existing agent setup that already uses Ornith 1.0
  • Running a 9B or 35B coding agent locally, with GGUF builds for llama.cpp-style tools
  • Comparing the gain from Ornith 1.0 to 1.5 in your own harness

Watch out for

  • Ornith 1.5 supersedes it with the same sizes and higher self-reported scores
  • The announcement lists a 31B dense model, but no 31B checkpoint is published in Ornith's Hugging Face organization
  • The 397B card says it is designed for single-GPU deployment, which does not fit a file of about 794 GB
  • All benchmark results are Ornith's own, and the blog's 35B score of 64.4 differs from the 64.2 in the model card

Performance claims come from the developer and have not been independently verified.

How much memory does it need?

Estimates for the Ornith-1.0-9B variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.

Variants Parameters BF16 FP8 / INT8 4-bit (Q4) Fits on
Ornith-1.0-9B BF16 9.41B 22 GB 11 GB 6.49 GB Laptop or phone
Ornith-1.0-35B BF16 35.1B 3B active 81 GB 40 GB 24 GB Gaming GPU
Ornith-1.0-35B (FP8) FP8 35.1B 3B active 81 GB 40 GB 24 GB Gaming GPU
Ornith-1.0-397B BF16 397B 17B active 913 GB 456 GB 274 GB Big-memory desktop
Ornith-1.0-397B (FP8) FP8 397B 17B active 913 GB 456 GB 274 GB Big-memory desktop

Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software.

VRAM calculator

The license in plain words

The shortest and simplest permissive license. You can do almost anything with the model as long as you keep the copyright and license notice.

More about MIT

Frequently asked questions

What is Ornith 1.0?
Ornith 1.0 is the first release of DeepReinforce's coding-agent models, in 9B, 35B and 397B sizes under the MIT license. The company has since replaced it with Ornith 1.5, which has the same sizes and higher scores, but the 1.0 files are still downloaded a lot.
Can I use Ornith 1.0 commercially?
Yes. Ornith 1.0 is released under MIT, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
Is Ornith 1.0 open source?
Ornith 1.0 is released under MIT, a license approved by the Open Source Initiative. Training data and training code may not be public, so it is best described as open weights under an open-source license.
Can I run Ornith 1.0 on my own computer?
Yes. The smallest variant, Ornith-1.0-9B, needs about 6.49 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
When was Ornith 1.0 released?
Ornith 1.0 was released by DeepReinforce on June 2026.
How long a context window does Ornith 1.0 have?
Ornith 1.0 supports a context window of up to 256K tokens.

Official links

Sources

Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.

Release date and facts last verified November 11, 2026.

Related models

Model Released Parameters Context License Smallest variant needs
Ornith 1.5 DeepReinforce 9.65B – 403B 17B active 256K MIT (Permissive) 6.66 GB Laptop or phone
Qwen3.5 Alibaba Qwen 873M – 403B 17B active 256K Apache 2.0 (Permissive) 0.602 GB Laptop or phone
Gemma 4 Google DeepMind 5.1B – 30.7B 256K Apache 2.0 (Permissive) 3.52 GB Laptop or phone
Qwen3.6 Alibaba Qwen 27.8B – 36B 3B active 256K Apache 2.0 (Permissive) 19 GB Gaming GPU