Capabilities
Minimum system memory
Tags
Last updated
Updated 5 hours agobyREADME
Parable models are trained on real agent work: multi-step tool use, planning, and thinking traces captured from Claude Fable 5 and GPT-5.5 sessions, not synthetic Q&A. Every release is eval-gated against its base before it ships.
This page hosts the Qwen3 line of the family. It adds structured <think> reasoning and agent-style task execution to Qwen3, tuned for local coding and terminal work.
Highlights:
<think> reasoning from actual Claude Fable 5 and GPT-5.5 sessions, not synthetic Q&A.Sources
The underlying model files this model uses
Based on
GGUF
Evaluation (held-out test split, identical eval code for base and fine-tune):
Token accuracy 4b: 0.683 → 0.782. Strictly graded qualitative results are published on the Hugging Face cards.
Usage notes
<think>...</think> block before the final answer.Family
parable/granite4.1-fablelms get)ollama run parable/qwen3-fable, plus the sibling line and the family flagship parable/fableBase model: Qwen3 (Apache-2.0). Training data: Glint-Research/Fable-5-traces (AGPL-3.0), gpt5.5-terminal (MIT). Third-party assistant traces; providers' terms may apply to downstream training.