DeepReinforce Releases Ornith-1.0, Open-Source Agentic Coding Model in 9B to 397B Sizes
DeepReinforce has released Ornith-1.0, an MIT-licensed model designed for agentic coding tasks with variants ranging from 9B to 397B parameters. Built on top of Apache 2.0-licensed Gemma 4 and Qwen 3.5 base models, the company claims it achieves state-of-the-art performance among open-source models of comparable size on coding benchmarks.
DeepReinforce Releases Ornith-1.0, Open-Source Agentic Coding Model in 9B to 397B Sizes
DeepReinforce has released Ornith-1.0, an MIT-licensed model designed for agentic coding tasks with variants ranging from 9B to 397B parameters. The model is available in four configurations: 9B Dense, 31B Dense, 35B MoE (Mixture of Experts), and 397B MoE.
Technical Foundation
Ornith-1.0 is built on top of pretrained Gemma 4 and Qwen 3.5 base models, both of which carry Apache 2.0 licenses. This represents a notable shift from earlier Gemma models, which were encumbered by additional Terms of Use restrictions. The Apache 2.0 licensing on both base models makes the derivative work's MIT license legally compatible.
DeepReinforce claims the model achieves state-of-the-art performance among open-source models of comparable size on coding benchmarks, though specific benchmark scores have not been disclosed.
Deployment and Performance
The model is available as GGUF quantized weights on Hugging Face. The 35B variant in Q4_K_M quantization weighs 20GB and can run locally via LM Studio. Early testing shows the model handles multi-step agentic workflows competently.
In documented testing against a Datasette codebase, Ornith-1.0 successfully navigated complex code search queries including "find the code that decodes the actor cookie" and "find the code that opens the insert dialog when the button is clicked." The model demonstrated the ability to chain multiple tool calls to complete these tasks.
Performance testing on consumer hardware shows the 35B quantized model running at 103 tokens per second for generation tasks.
Company Background
DeepReinforce appears to be a new entrant in the AI model space. The earliest available research from the organization is a June 2025 paper titled "CUDA-L1: Improving CUDA Optimization via Contrastive Reinforcement Learning." No prior model releases from DeepReinforce have been documented.
Availability
All Ornith-1.0 variants are available immediately with open weights under the MIT license. GGUF quantized versions are distributed through Hugging Face for local deployment.
What This Means
Ornith-1.0 represents a significant addition to open-source agentic coding models, particularly for developers seeking local deployment options. The MIT licensing and compatibility with existing Apache 2.0 base models removes common licensing friction. However, without disclosed benchmark scores or extensive independent testing, claims of state-of-the-art performance among comparable models remain unverified. The availability of multiple size variants, including the computationally efficient MoE architectures, provides deployment flexibility across different hardware constraints.
Related Articles
Qwen4 Max launches with 1M-token context and $1.50/$4.50 per 1M token pricing
Qwen4 Max, the flagship model in Alibaba's Qwen4 series, is listed on OpenRouter with a 1M-token context window. It accepts text, image and video input, returns text, and is priced at $1.50 per 1M input tokens and $4.50 per 1M output tokens.
Runway releases Praxis-1 robot model with open weights under Apache 2.0, tops World Action Models on RoboLab-120
Runway has released Praxis-1, a robot control model, with open weights and source code under the Apache 2.0 license. It scores 46.2 percent on Nvidia's RoboLab-120 leaderboard, ahead of FLUX 3 Action at 42.9 percent, in tests Runway ran itself.
Translated releases Lara Mini, a 7.2B-parameter MoE research model with 600M active parameters
Translated announced Lara Mini on November 12, 2026, a research model trained from scratch on 6 trillion tokens across English and 44 other languages, including all 24 official EU languages. The mixture-of-experts system has 7.2 billion parameters and activates 600 million per token.
Endless Frontier open-sources Fx-Bio-1031, a 125B MoE bioinformatics model with 512K context
Endless Frontier (无界前沿) has open-sourced Fx-Bio-1031, a bioinformatics model with 125B total parameters and 6B active per token, with a 512K context window. The company claims 70.59% Pass@5 on the Difficult subset of Anthropic's BioMysteryBench, against 47.06% for GPT-6-Astra and 52.94% for Claude Opus-5.
Comments
Loading...