flashback2k/FlashModel-Qwen-3.6-35B-A3B-Maxed

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:2Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 9, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

FlashModel-Qwen-3.6-35B-A3B-Maxed by flashback2k is a 35.1 billion parameter, bilingual (English/Russian) fine-tune of Qwen3.6-35B-A3B, designed for advanced reasoning, coding, and agentic tool use. This model features a unique controllable reasoning effort mechanism, allowing users to specify reasoning depth from 'low' to 'max' in the system prompt. It excels in code generation across multiple languages and supports agentic operations like file I/O and shell commands, making it suitable for complex development tasks.

Loading preview...

Overview

FlashModel-Qwen-3.6-35B-A3B-Maxed is a 35.1 billion parameter, bilingual (English/Russian) fine-tune of the Qwen/Qwen3.6-35B-A3B model, developed by flashback2k. This model is specifically enhanced for advanced reasoning, coding, and agentic tool use, building upon the robust foundation of the Qwen architecture. It retains the MTP head for speculative decoding, ensuring efficient inference.

Key Capabilities

  • Controllable Reasoning Effort: Users can specify reasoning depth (low, medium, high, xhigh, max) directly in the system prompt, providing granular control over the model's thought process. This is based on matched-pairs data, ensuring behavioral rather than artificial length-based control.
  • Extensive Code Generation: Proficient in generating code for Rust, TypeScript, JavaScript, Python, QML, HTML, CSS, and Bash.
  • Agentic Tool Use: Capable of performing operations such as file read/write, shell commands, and Git operations.
  • Bilingual Proficiency: Supports English and Russian dialogue, with approximately 22% Russian data included in its training.

Good For

  • Complex Reasoning Tasks: Ideal for applications requiring detailed, multi-step reasoning where the depth of thought can be dynamically adjusted.
  • Software Development: Excellent for code generation, debugging assistance, and automating development workflows across various programming languages.
  • Agentic Applications: Suitable for building intelligent agents that interact with file systems, execute shell commands, or manage Git repositories.
  • Bilingual Chatbots: Effective for conversational AI systems that need to operate in both English and Russian, particularly in technical or problem-solving contexts.

Finetunes
2 models