SL-AI/GRaPE-2.5-Helios

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 13, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

SL-AI/GRaPE-2.5-Helios is a 9 billion parameter multimodal language model developed by SLAI (Skinnertopia Lab for Artificial Intelligence) with a 32768 token context length. It supports image and text inputs, producing text outputs, and features controllable thinking modes and work efforts for agentic tasks. This model is optimized for creative tasks, agentic coding, and STEAM, aiming to provide a less robotic prose and enhanced generalization compared to previous generations.

Loading preview...

GRaPE 2.5 Helios: Multimodal Agentic LLM

GRaPE 2.5 Helios is a 9 billion parameter multimodal language model from the third-generation GRaPE family, developed by SLAI (Skinnertopia Lab for Artificial Intelligence). It accepts image and text inputs to generate text outputs, and is designed for generalistic tasks in local environments. This model is a successor to GRaPE 2.1 Flash, featuring substantially improved training data focused on quality over quantity, and aims to overcome shortcomings of previous generations with a stronger, more diverse training corpus and a less robotic prose style.

Key Capabilities

  • Multimodal Input: Processes both image and text inputs.
  • Controllable Reasoning: Features adjustable thinking modes (minimal, low, medium, high, xtra-hi) to control reasoning depth, placed at the end of the prompt.
  • Agentic Work Effort: Offers controllable work effort modes (minimal, low, medium, high, xtra-hi) for tool-based tasks, also placed at the end of the prompt.
  • Enhanced Creativity: Specifically trained on creative conversations and story writing to produce unique and less "AI-like" responses.
  • Multilingual Thinking: Capable of thinking in languages other than English, with support for thought summaries.
  • Idea Generation: Scaffolds entire thinking blocks and generates multiple ideas to prevent "what if" loops and improve task focus.
  • Optimized for: Creative Tasks, Agentic Coding, and STEAM (Science, Technology, Engineering, Arts, and Mathematics).

Good for

  • Complex Agentic Tasks: Especially those requiring multi-step reasoning and tool use, benefiting from preserved thinking blocks in context.
  • Creative Content Generation: Ideal for applications needing unique stories, conversations, or design concepts that avoid typical LLM-isms.
  • Multilingual Interactions: When seamless conversations and thinking in various languages are required.
  • Applications Requiring Controllable Output: Users can fine-tune the model's reasoning depth and work effort for specific task requirements.

Finetunes
2 models