🤖 Release of the Qwen3.6-34B-80L-Fable-5-Heretic Model
The Qwen3.6-34B-80L-Fable-5-Heretic model has been introduced, which is a distillation of Fable-5 agent trajectories based on the Qwen3.6-27B architecture. By increasing the number of layers from 64 to 80, the model has reached 34 billion parameters and improved CoT (Chain-of-Thought) reasoning capabilities.
🌍 The use of hybrid attention and MTP weights increases throughput when working with a long context of up to 256K tokens, which is critical for AI agents.
👤 The model allows for running powerful logical systems locally. Thanks to optimization for vLLM and support for speculative decoding, it runs approximately 2x faster than its analogs.
Source 1: https://huggingface.co/hiebo/Qwen3.6-34B-80L-Fable-5-Heretic