Ornith-1.5-397B-GGUF

Chirp Chirp! 🐦 We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement. Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a fixed set of human-curated tasks and manually designed harnesses, Ornith-1.5 continuously generates new training tasks, discovers effective strategies for solving them, and improves the policy through reinforcement learning. For more details on the task, harness, and rollout reward design, please refer to our [blog](https://ornith.ai/ornith_1_5.html). This model card documents **Ornith-1.5-397B**, the flagship member of the Ornith-1.5 family — a 397B mixture-of-experts model. It scores 86.1 on Terminal-Bench 2.1 and 56.0 on DeepSWE, performing on par with Claude Opus 4.8 (85.0 and 59.0) while outperforming leading open-source models of similar scale, including GLM-5.2 and DeepSeek-V4-Flash-0731. * All results reported for Ornith-1.5 are averaged over five independent runs.<br/> * Terminal-Bench 2.1 (Terminus-2): We evaluate Terminal-Bench 2.1 using the Harbor/Terminus-2 framework with parser=json, temperature=1.0, top_p=1.0, and a 128K context window. Each run uses a 4-hour timeout with 32 CPU cores and 48GB RAM, and results are averaged over 5 runs. We adjust the Qwen chat template to ensure consistency between training and inference (https://huggingface.co/ornith-ai/Ornith-1.5-397B/blob/main/chat_template.jinja), and modify Harbor to align with vLLM's reasoning_content key.<br/> * Terminal-Bench 2.1 (Claude Code): We evaluate Terminal-Bench 2.1 using Claude Code 2.1.126 with parser=json, temperature=1.0, top_p=1.0, max_new_tokens=131072. Results are averaged over 5 runs. Again, Qwen chat template needs to be modified.<br/> * SWE-Bench Verified, Pro and Multilingual: using OpenHands harness with temp=1.0, top_p=0.95, 256k context window. Anti-hacking safeguards are applied throughout evaluation: Git history is removed from the local repository image to prevent access to prior solutions or commits; network access is disabled, preventing the model from retrieving external information or resources.<br/> * DeepSWE: Evaluated using the Claude Code harness with temperature=1.0, top_p=0.95, and a 256K context window.<br/> * SWE Atlas QnA: using mini SWE agent harness with temp=1.0, top_p=0.95, 128K context window. Results are averaged over 5 runs.<br/> * NL2Repo: with temperature=1.0, top_p=1.0, 400K context, 48K output. Access to specified GitHub repositories and pip packages is blocked to prevent reward hacking.<br/> * HLE: Evaluated using Claude 4.6 Opus as the judge model.<br/> * MCP-Atlas: All models were evaluated in thinking mode on the 500-task public subset, with a 10-minute timeout per task. We use Claude 4.8 Opus as the judge model.<br/> * Toolathlon-Verified: We use the official evaluation service with the maximum token limit set to 128K.<br/> * ClawEval: An agentic code benchmark over real-user task distributions; temp=0.6 and 256K context.<br/>

Other versions in the Ornith family