nlpguy/Qwen3.8-27B-Fimi-3

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 24, 2026Architecture:Transformer0.0K Featherless Exclusive Cold

The nlpguy/Qwen3.8-27B-Fimi-3 is a 27 billion parameter language model based on the Qwen architecture, created by nlpguy through a merge of several pre-trained Qwen models. This model was developed using the Reinforced Agent Merging Plus (Tensor-Local) method, with Qwen/Qwen3.5-27B as its base. It is specifically noted for its potential in agentic performance, building upon its predecessor, Fimi 2.

Loading preview...

Model Overview

nlpguy/Qwen3.8-27B-Fimi-3 is a 27 billion parameter language model built upon the Qwen architecture. It was created by nlpguy using the mergekit tool, specifically employing the Reinforced Agent Merging Plus (Tensor-Local) method. The base model for this merge was Qwen/Qwen3.5-27B.

Key Characteristics

  • Architecture: Based on the Qwen series of models.
  • Parameter Count: 27 billion parameters.
  • Merge Method: Utilizes the advanced Reinforced Agent Merging Plus (Tensor-Local) technique, which is detailed in this paper.
  • Component Models: This model is a composite of seven distinct Qwen-based models, including contributions from DavidAU, beyoru, empero-ai, bottlecapai, and TeichAI, merged with the Qwen/Qwen3.5-27B base.

Intended Use

Fimi 3 is designed to offer enhanced capabilities, particularly in agentic performance. Users seeking strong agentic behavior are encouraged to consider this model, with a note that Fimi 2 is also recommended for similar agentic tasks.