SL-AI/GRaPE-2.1-Flash

VISIONConcurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Apr 19, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

SL-AI/GRaPE-2.1-Flash is a 9 billion parameter multimodal language model developed by the Skinnertopia Lab for Artificial Intelligence (SLAI), built on the Qwen3.5 architecture. It accepts image and text inputs to produce text outputs, featuring an extended thinking mode system for controllable reasoning depth. This model is specifically post-trained with a heavy emphasis on code, STEAM, logical reasoning, and structured problem-solving, making it suitable for advanced device deployment and complex analytical tasks.

Loading preview...

GRaPE 2.1 Flash: Multimodal Reasoning Model

GRaPE 2.1 Flash is the flagship mid-sized model from the second-generation GRaPE family, developed by the Skinnertopia Lab for Artificial Intelligence (SLAI). Built upon a Qwen3.5 9B base, it supports multimodal inputs (image + text) and generates text outputs. This model is a direct successor to GRaPE Flash, incorporating improved training data and a more capable foundation.

Key Capabilities & Improvements

  • Stronger Base Model: Utilizes Qwen3.5 9B, a significant upgrade from the previous Qwen3 VL base.
  • Expanded Thinking Modes: Features six discrete reasoning tiers (minimal, low, medium, high, xtra-Hi, auto) for controllable reasoning depth, activated via a <thinking_mode> tag at the end of prompts.
  • Proprietary Training Data: Post-trained on a curated, closed-source dataset with a strong focus on:
    • Code (~50% of post-training data)
    • STEAM (Science, Technology, Engineering, Arts, and Mathematics)
    • Logical reasoning and structured problem-solving
  • Increased Parameters: At 9 billion parameters, it offers enhanced intelligence compared to its 7B predecessor.

When to Use GRaPE 2.1 Flash

This model is particularly well-suited for:

  • Complex coding tasks requiring deep analysis.
  • Multi-step mathematical problems.
  • Deep analytical work and structured problem-solving.
  • Agentic use cases where controlled reasoning depth is beneficial.

Benchmarks for GRaPE 2.1 Flash are currently underway and will be published upon completion. The model is designed to punch above its weight class on structured reasoning tasks while remaining deployable on consumer hardware.

Finetunes
3 models