Ornith-1.0 Released: A New King of Agentic Coding Ascends, Open-Sourced under MIT

The Ornith-1.0 family of open-source coding models was released under the MIT license, achieving state-of-the-art results on agentic coding benchmarks including SWE-Bench Verified (82.4) and Terminal-Bench 2.1 (77.5), surpassing closed-source models. The four models range from 9B to 397B parameters, built on Gemma 4 and Qwen 3.5, and are fully deployable locally.

Member-only story Ornith-1.0 Released: A New King of Agentic Coding Ascends, Open-Sourced under MIT When it comes to coding, open source models have evolved from catching up to taking the lead. Not an exaggeration. Not clickbait. Early this morning, a model family named Ornith-1.0 was released as open source, sweeping all available benchmarks in the most hardcore arena—Agentic Coding. SWE-Bench Verified: 82.4. Terminal-Bench 2.1: 77.5. SWE-Bench Pro: 62.2. NL2Repo: 48.2. ClawEval: 77.1. Not "close to closed-source levels." It's "closed-source models haven't even publicly reached this score yet." When it comes to coding, open source models have gone from chasing to leading. And this time, it's a lead across the full parameter range, fully open-sourced under MIT, fully deployable locally. They even provide a GGUF version. Four Models, One Ambition Ornith-1.0 is not a single model; it's a family. Four specifications covering everything from a laptop to a server cluster. 9B Dense. 31B Dense. 35B MoE. 397B MoE. The smallest 9B can run on consumer-grade GPUs. The largest 397B MoE is aimed at enterprise-level private deployment. It is post-trained on top of Gemma 4 and Qwen 3.5—standing on the shoulders of giants and…