KIEFERSA/Sophea-Titan-1

VISIONConcurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jul 12, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

KIEFERSA/Sophea-Titan-1 is a 27 billion parameter multimodal chat model developed by Kiefer SA, fine-tuned from Qwen3.6-27B. It is optimized for Greek conversational tasks, offering formal and informal register control while retaining strong English language and vision capabilities. This model excels as a general-purpose Greek conversational assistant and for Greek-knowledge QA.

Loading preview...

Sophea-Titan-1: Greek Multimodal Chat Model

Sophea-Titan-1 is a 27 billion parameter multimodal chat model developed by Kiefer SA, built upon the Qwen3.6-27B architecture. It has been specifically fine-tuned for Greek language conversational tasks, incorporating register control and a fixed assistant identity. Notably, it maintains strong English language proficiency and multimodal vision capabilities.

Key Capabilities

  • Primary Language Focus: Optimized for general-purpose Greek conversation and Greek-knowledge question answering.
  • Multimodal: Retains the vision capabilities of its base model, Qwen3.6-27B.
  • Register Control: Offers the ability to switch between formal and informal conversational styles.
  • Performance: Achieves the strongest general Greek benchmark scores among comparable open models, outperforming its base model and other Greek LLMs like Krikri-8B. It also demonstrates excellent English retention and strong vision performance on MMStar benchmarks.
  • Efficiency: When run in "non-thinking" mode (recommended), it provides high-quality Greek output and can approach the performance of much larger frontier MoE models in Greek benchmarks.

Good For

  • Developers building Greek-centric conversational AI applications.
  • Applications requiring multimodal understanding in Greek.
  • Use cases where controlled conversational register is beneficial.

It is important to note that for optimal Greek output quality, the model should be served with enable_thinking=false in the chat template. The model is not evaluated for safety-critical or legal/medical decisions.