bugrabilge/Omni-31B-Turkish-Reasoning-Model

VISIONPricing:Input $0.48 / Cached $0.1 / Output $1.44Concurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:May 18, 2026License:gemmaArchitecture:Transformer0.0K Featherless Exclusive Cold

The Omni-31B-Turkish-Reasoning-Model by bugrabilge is a 31 billion parameter Gemma 4-based model fine-tuned for Turkish Chain-of-Thought (CoT) reasoning. It specializes in generating step-by-step explanations, multi-step problem-solving, and long-form analytical responses in Turkish. With a 32768 token context length, this model is optimized for structured, detailed answers in academic and explanatory contexts.

Loading preview...

Omni-31B Turkish Reasoning Model Overview

Omni-31B is a 31 billion parameter model developed by bugrabilge, built upon the google/gemma-4-31B base architecture. It has been extensively fine-tuned using 249K filtered Turkish Chain-of-Thought (CoT) data to excel in reasoning, explanatory analysis, and multi-step problem-solving in Turkish. The model is designed to produce visible, step-by-step reasoning within <think> blocks before delivering a clear, structured answer to the user.

Key Capabilities

  • Turkish Chain-of-Thought Reasoning: Generates detailed, step-by-step thought processes within <think> blocks for planning, intermediate verification, and structured response generation.
  • Explanatory Analysis: Provides long-form Turkish explanations across various subjects like science, history, literature, and philosophy.
  • Multi-step Reasoning: Breaks down complex questions into manageable steps for comprehensive answers.
  • Format Discipline: Produces clean, user-facing answers following the internal reasoning block.
  • Long-form Output: Optimized to generate academic-style responses typically ranging from 5,000 to 7,000 tokens.

Recommended Use Cases

  • Turkish Educational Assistant: Ideal for generating explanatory content and detailed analyses.
  • Academic Content Creation: Suitable for drafting academic-style Turkish content and long-form analyses in various fields.
  • Research on Turkish Reasoning: A valuable tool for exploring and developing Turkish reasoning capabilities in AI.

Technical Details

The model was fine-tuned using Full Supervised Fine-Tuning with 16-bit weights and supports a maximum training sequence length of 8192 tokens. It is a text-only model, despite the base architecture potentially supporting multimodal inputs, as its training was exclusively text-based. Users should note that the model tends to produce long, detailed answers and requires explicit system prompts for more concise outputs or consistent <think> block triggering.

Finetunes
1 models