---
format: "aidr-story-markdown/v1"
id: "474551e804f1b969b50f4471f7c8ba496841b28a45c894fea3281743a17dfded"
canonical_url: "https://aidr.today/474551e8?lang=en"
title: "Ornith 397B Open Model Beats Claude Opus 4.8, Matching Top Closed AI in Coding"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-08-19T16:59:17.000Z"
category: "Models"
topics: ["open-source","llm","coding","reasoning","swe-bench"]
source_urls: ["https://huggingnews.com/ai/ornith-397b-open-model-beats-claude-opus-48-matching-top-closed-ai-in-co-6ce5b0fb","https://x.com/ornith_/status/2090074077084127302","https://x.com/ornith_/status/2090075048812118352","https://x.com/ornith_/status/2090078493640171908","https://x.com/ornith_/status/2090078992267448583","https://x.com/rohanpaul_ai/status/2090085614989533532","https://x.com/testingcatalog/status/2090088730203685064","https://x.com/ornith_/status/2090258105682854390"]
summary: "Ornith has released the Ornith-1.5 family of open source large language models featuring 9B Dense, 35B Mixture of Experts (MoE), and 397B MoE variants. The 397B model outperformed Claude Opus 4.8 on four reasoning and coding benchmarks, including Terminal-Bench 2.1 and SWE-bench Verified, by using a reinforcement learning loop that generates its own tasks and grading scaffolds. All variants are distributed under the MIT License for unrestricted commercial and research use. The launch includes a 9B-Mobile version quantized to 1.5 GB for deployment on iPhone and Android devices. The 35B MoE model activates only 3B parameters per token while exceeding the performance of Google's Gemma 4-31B and Meta's Muse Glimmer-30B. The series is available in FP8, NVFP4, GGUF, and MLX formats for integration with platforms such as Ollama, LM Studio, and Unsloth."
---

# Ornith 397B Open Model Beats Claude Opus 4\.8, Matching Top Closed AI in Coding

> [Open the canonical story](<https://aidr.today/474551e8?lang=en>)

**Published:** 2026-08-19T16:59:17.000Z
**Category:** Models
**Topics:** open\-source, llm, coding, reasoning, swe\-bench

## Summary

Ornith has released the Ornith\-1\.5 family of open source large language models featuring 9B Dense, 35B Mixture of Experts \(MoE\), and 397B MoE variants\. The 397B model outperformed Claude Opus 4\.8 on four reasoning and coding benchmarks, including Terminal\-Bench 2\.1 and SWE\-bench Verified, by using a reinforcement learning loop that generates its own tasks and grading scaffolds\. All variants are distributed under the MIT License for unrestricted commercial and research use\. The launch includes a 9B\-Mobile version quantized to 1\.5 GB for deployment on iPhone and Android devices\. The 35B MoE model activates only 3B parameters per token while exceeding the performance of Google's Gemma 4\-31B and Meta's Muse Glimmer\-30B\. The series is available in FP8, NVFP4, GGUF, and MLX formats for integration with platforms such as Ollama, LM Studio, and Unsloth\.

## Sources

- [Story source](<https://huggingnews.com/ai/ornith-397b-open-model-beats-claude-opus-48-matching-top-closed-ai-in-co-6ce5b0fb>)
- [Story source](<https://x.com/ornith_/status/2090074077084127302>)
- [Story source](<https://x.com/ornith_/status/2090075048812118352>)
- [Story source](<https://x.com/ornith_/status/2090078493640171908>)
- [Story source](<https://x.com/ornith_/status/2090078992267448583>)
- [Supporting source](<https://x.com/rohanpaul_ai/status/2090085614989533532>)
- [Supporting source](<https://x.com/testingcatalog/status/2090088730203685064>)
- [Story source](<https://x.com/ornith_/status/2090258105682854390>)