TendieLabs/Frank-26B-A4B

VISIONConcurrent Unit Cost:2Model Size:26BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 7, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

TendieLabs/Frank-26B-A4B is a 25.2 billion parameter Mixture-of-Experts (MoE) model, fine-tuned from Google's Gemma 4 26B-A4B-it, with approximately 3.8 billion active parameters per forward pass. Optimized for code generation, debugging, and code review in C# and Python, it features learned task discrimination for 10x faster inference on coding tasks while retaining full reasoning capabilities. This model is designed for local deployment as a coding assistant, particularly in environments requiring US/NATO-aligned model origins.

Loading preview...

Frank 26B-A4B: A Code-Optimized Gemma 4 Fine-tune

Frank 26B-A4B is a fine-tuned version of Google's Gemma 4 26B-A4B-it by TendieLabs, specifically engineered for coding tasks. This Mixture-of-Experts (MoE) model boasts 25.2 billion total parameters, with an efficient ~3.8 billion active per forward pass, offering deep knowledge at near-4B speed.

Key Capabilities & Differentiators

  • Learned Task Discrimination: Frank intelligently skips unnecessary reasoning on straightforward code generation, achieving up to 10x faster inference on coding tasks compared to the base Gemma 4, while maintaining full reasoning for complex problems like debugging or code review.
  • NATO-Aligned: Built on a US-origin base model with an Apache 2.0 license, making it suitable for deployment in environments with strict supply chain policies.
  • Optimized for Code: Fine-tuned on ~33K coding examples across C#, Python, multi-turn debugging, and general reasoning.
  • Stable Thinking Mode: Includes Google's latest chat template patches to resolve known 'ghost thought channel' issues in Gemma 4, ensuring stable chain-of-thought reasoning.
  • Performance: Benchmarks show Frank matches base Gemma 4's quality on VBA-to-Python conversion while being significantly faster. It also retains strong reasoning capabilities across various complex tasks.

Recommended Use Cases

  • Local Coding Assistant: Ideal for use with tools like OpenCode or any OpenAI-compatible client for on-premises development.
  • Code Generation, Debugging, and Review: Excels in C# and Python programming tasks.
  • VBA-to-Python Conversion: Proven effective for converting VBA reports to Python.
  • Environments with Compliance Needs: Suitable for organizations requiring US/NATO-aligned AI model origins.

Finetunes
2 models