The AI model 'Ornith-1.5' has been released, an open model that covers everything from large models like the Claude Opus 4.8 to lightweight models capable of running on smartphones.

The AI research organization Ornith released its AI model ' Ornith-1.5 ' on August 19, 2026. Ornith-1.5 is released in three separate models: ' Ornith-1.5-397B ,' ' Ornith-1.5-35B-A3B, ' and ' Ornith-1.5-9B .' The largest model, Ornith-1.5-397B, has achieved benchmark scores equivalent to those of the Claude Opus 4.8. Furthermore, the quantized version of Ornith-1.5-9B is said to be executable on smartphones.

Ornith-1.5: From Self-Scaffolding to Self-Improvement | Ornith Blog

Aloha! 🌺Introducing Ornith-1.5, a family of open-source LLMs spanning 9B Dense, 35B MoE, and 397B MoE, trained with self-improving strategies.

It achieves state-of-the-art performance among open-source models of comparable size and delivers comparable performance to Claude Opus… pic.twitter.com/IzR6eBNpZ8

— Ornith (@ornith_) August 19, 2026

Ornith-1.5 incorporates a self-improvement loop into its learning process, enabling autonomous performance improvement by repeatedly creating a loop in which the AI model itself proposes more difficult tasks during the learning process.

Ornith-1.5-397B is a MoE model with a total of 397 billion parameters. The benchmark results for 'Ornith-1.5-397B', 'Ornith-1.0-397B', 'DeepSeek-V4-Flash-0731', 'GLM-5.2', and 'Claude Opus 4.8' are shown below. Ornith-1.5-397B outperforms Claude Opus 4.8 in several tests.

Ornith-1.5-35B-A3B is a MoE model with a total of 35 billion parameters and 3 billion active parameters. Below are the performance comparison results for 'Ornith-1.5-35B-A3B', 'Ornith-1.0-35B-A3B', 'Qwen3.6-35B-A3B', 'Muse-Glimmer-30B', and 'Gemma-4-31B'. Despite being a MoE model, Ornith-1.5-35B-A3B outperforms the Dense models Muse-Glimmer-30B and Gemma-4-31B.

Ornith-1.5-9B is a Dense model with 9 billion parameters. The following are the results of a comparison between 'Ornith-1.5-9B', 'Ornith-1.0-9B', 'Qwen3.5-9B', 'Qwen3.6-35B-A3B', and 'Gemma-4-31B'. Ornith-1.5-9B outperforms Gemma-4-31B, which has a larger number of parameters, in most tests. There is also a quantized version called 'Ornith-1.5-9B-Mobile' intended for use on smartphones.

Ornith-1.5 is released as an open model under the MIT License. Various models are distributed via Hugging Face, and a quantized version by Ornith was also released on the same day.

Ornith-1.5 - a ornith-ai Collection
https://huggingface.co/collections/ornith-ai/ornith-15