teddy19032/Ornith-1.5-9B-heretic

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 29, 2026License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Cold

teddy19032/Ornith-1.5-9B-heretic is a 9 billion parameter language model, a decensored version of ornith-ai/Ornith-1.5-9B created using the Heretic v1.2.0 Arbitrary-Rank Ablation (ARA) method. It maintains a 32768 token context length and is optimized for reasoning and agentic tasks, demonstrating improved performance in coding and problem-solving benchmarks compared to its original counterpart. This model is designed for efficient single-GPU deployment and excels in scenarios requiring reduced refusal rates.

Loading preview...

Ornith-1.5-9B-heretic: A Decensored, Reasoning-Optimized Model

This model is a 9 billion parameter variant, derived from the ornith-ai/Ornith-1.5-9B model through a decensoring process using the Heretic v1.2.0 Arbitrary-Rank Ablation (ARA) method. It significantly reduces refusal rates from 99/100 to 18/100 while maintaining a low KL divergence of 0.0398 compared to the original model.

Key Capabilities

  • Enhanced Reasoning: The model is designed as a reasoning model, by default opening assistant turns with a <think> ... </think> block before the final answer, with a reasoning parser to separate chain-of-thought.
  • Improved Coding Performance: Outperforms the original Ornith-1.5-9B on various coding benchmarks, including Terminal-Bench 2.1 (Terminus-2: 46.2 vs 43.1, Claude Code: 47 vs 40.6), SWE-bench Verified (70.6 vs 69.4), SWE-bench Pro (47.5 vs 42.9), and NL2Repo (32.4 vs 27.2).
  • Agentic Functionality: Supports OpenAI-compatible tool calling, making it suitable for integration with standard agent frameworks like Ollama, Atomic.chat, llama.cpp, Hermes Agent, and OpenClaw.
  • Long Context Handling: Features a native context window of 262,144 tokens, extendable to approximately 1M tokens using YaRN (Yet another RoPE Normalization) scaling.

Good for

  • Agentic Workflows: Ideal for developers building agents that require tool use and complex reasoning.
  • Coding and Software Engineering Tasks: Excels in code generation, understanding, and problem-solving, particularly in terminal-based coding environments.
  • Applications Requiring Reduced Refusals: Suitable for use cases where a less restrictive model response is desired, due to its decensored nature.

Finetunes
2 models