Smilyai-labs/Mira-1-large

TEXT GENERATIONPricing:Input $0.48 / Output $0.96Concurrent Unit Cost:1Model Size:14BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 16, 2026License:otherArchitecture:Transformer0.0K Featherless Exclusive Cold

Mira-1-large is a 14.8 billion parameter conversational and reasoning model developed by SmilyAI, built upon the Qwen3-14B architecture via the Nova-2-Large lineage. It features a 40,960-token context length and is specifically fine-tuned for combining strong coding and mathematical capabilities with a distinctive, friendly personality. This model excels at structured reasoning, debugging, and technical discussion, making it suitable for AI/ML experimentation and programming assistance.

Loading preview...

SmilyAI's Mira-1-large: A Specialized Conversational and Reasoning Model

Mira-1-large, developed by SmilyAI, is a 14.8 billion parameter model with a 40,960-token context window, designed for conversational AI, coding, and reasoning. It is the final model in the Nova series, built through a multi-stage lineage starting from Qwen3-14B, incorporating a DARE-TIES merge for enhanced code and mathematics capabilities in its immediate base, Nova-2-Large. Mira is then fine-tuned using LoRA to develop a unique personality, refined conversational and reasoning behaviors, and improved code-related interaction.

Key Capabilities

  • Reasoning: Fine-tuned for structured problem-solving, breaking down complex tasks, checking intermediate conclusions, and methodical debugging. It can produce reasoning traces using <think> tags when prompted appropriately.
  • Coding: Inherits strong coding abilities from its Nova-2-Large lineage, supporting Python, machine learning, AI engineering, debugging, and algorithmic reasoning. It focuses on reviewing and improving code rather than just generation.
  • Conversational Personality: Designed with a curious, friendly, energetic, and technically enthusiastic persona, acting as an AI engineering companion. It also features an optional "Roast Mode" for playful, constructive code review.

Good For

  • AI/ML experimentation and personal research
  • Programming assistance, code debugging, and technical brainstorming
  • Mathematics and technical discussions
  • Conversational experimentation where a distinct personality is desired

Limitations

Users should be aware that Mira can still hallucinate facts, generate incorrect code, make mathematical mistakes, and misinterpret ambiguous prompts. Its outputs require human review, especially for important tasks, and it is not intended for safety-critical applications without additional safeguards.

Finetunes
1 models