Smilyai-labs/Mira-1-XL

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 17, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

Smilyai-labs/Mira-1-XL is a 27 billion parameter multimodal (vision + text) model merge based on Qwen3.5-27B, designed for strong instruction following and improved reasoning. It integrates Claude-Opus-style reasoning and Fable-style creative distills, enhancing both analytical and narrative outputs. This model maintains Qwen3.5's vision capabilities, allowing for image+text conversations, and serves as a premium daily driver for general chat, long-form writing, and agent-style planning.

Loading preview...

Mira-1-XL: A Multimodal Daily Driver

Smilyai-labs/Mira-1-XL is a 27 billion parameter multimodal model, combining vision and text capabilities. Built upon the Qwen3.5-27B architecture, it integrates reasoning and creative distills from Claude-Opus and Fable-style models to offer a balanced and powerful instruction-following experience. The model maintains Qwen3.5's inherent vision capabilities, enabling image+text conversations.

Key Capabilities

  • Multimodal Interaction: Processes both image and text inputs to generate text outputs.
  • Enhanced Reasoning: Incorporates Claude-Opus-style reasoning for improved "thinking" behavior and structured solutions.
  • Creative Output: Blends Fable-style creative distills for better narrative generation and planning.
  • Strong Instruction Following: Designed as a premium "daily driver" for general chat and complex instructions.
  • BF16 Precision: Utilizes BF16 weights for efficient performance.

Intended Use Cases

Mira-1-XL is particularly well-suited for:

  • Multimodal Q&A (e.g., analyzing images, summarizing screenshots).
  • Long-form writing tasks, including stories, scripts, and outlines.
  • Agent-style planning and multi-step task decomposition.
  • General chat and instruction-following scenarios.
  • "Reasoning-ish" tasks where base Qwen models might be less effective.

It is important to note that while powerful, the model is not ideal for safety-critical domains without human review, does not guarantee factuality, and DARE-style merges can be stochastic unless RNG seeds are fixed.

Finetunes
1 models