lemuralabs/Qwen3.6-27B-Fable-Agentic

<p align="center"> <img src="logo.png" alt="Lemura Labs" width="110"/> </p>

<div align="center">

Qwen3.6-27B-Fable-Agentic

Agentic - Reasoning - Tool-Calling - Top-Tier Knowledge

Beats its base model on knowledge - and acts like an agent.

</div>

Qwen3.6-27B-Fable-Agentic is a powerful 27B agentic model that reasons step-by-step and operates tools across multi-turn sessions, while delivering top-tier general knowledge. Fine-tuned from Jackrong/Qwopus3.6-27B-v2.

Benchmarks

Evaluated head-to-head against the base model on identical questions, with an 8192-token budget so step-by-step reasoning is never truncated.

Knowledge improved, hard reasoning preserved: +1.43 over base on MMLU-Pro, matching it on GSM8K math and on GPQA-Diamond (graduate-level science, ~frontier-class for a 27B) - with agentic tool-calling added on top.

<details> <summary><b>MMLU-Pro by category</b></summary>

</details>

Highlights

• —Top-tier knowledge - 88.29% MMLU-Pro, above the base model
• —Strong math - 97.33% GSM8K
• —Agentic tool-calling across multi-turn sessions
• —Step-by-step reasoning built in

Capabilities

• —Agentic, multi-turn tool-calling
• —Chain-of-thought reasoning
• —Top-tier general knowledge and math
• —Multimodal (vision + text) architecture

Usage (vLLM)

vllm serve lemuralabs/Qwen3.6-27B-Fable-Agentic --trust-remote-code --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --max-model-len 16384

Tip: this model reasons before answering - give it a generous max_tokens so it can finish its chain of thought.

License

Apache 2.0