Skip to main content
Kaino.dev
Discover
Evals
News
Academics
Insights
Kaino.dev

Discover, evaluate, and compare AI tools, models, and agents.

Explore

  • Discover
  • Evaluations
  • News
  • Academics
  • Insights

Community

  • Twitter
  • YouTube
  • Instagram
Privacy PolicyTerms of Service

© 2026 Kaino.dev. All rights reserved.

Version 1.1.0
Discover/MODELS/Granite 4.0 H Small
Granite 4.0 H Small logo

MODELS

Granite 4.0 H Small

by IBM

modelibmgranite-4
Visit WebsiteDocumentationGitHub

Overview

IBM Granite-4.0-H-Small is a 32B total, 9B active hybrid MoE instruction model for enterprise RAG, agents, coding, tool use, and multilingual dialog.

Details

Granite-4.0-H-Small is an IBM Granite 4 language model described by IBM as a hybrid MoE model with 32B total parameters and 9B active parameters. IBM documentation lists it for RAG and agent use cases and shows watsonx.ai chat API usage with model_id ibm/granite-4-h-small. IBM watsonx.ai documentation identifies granite-4-h-small as an instruction-following Granite 4 model with a 131,072-token context window, tool use, RAG, function calling, FIM code, multilingual dialog, and API pricing tier details. The official IBM Granite Hugging Face model card states it was released October 2, 2025 under the Apache-2.0 license and highlights capabilities such as summarization, RAG, code, tool calling, and multilingual dialog.

When to Use

Use when evaluating an IBM Granite model for enterprise RAG or agent workflows including multi-tool agents and customer support automation described in IBM’s Granite 4.0 launch materials. Use when you need a long-context instruction-following model with documented support for tool use function calling FIM code multilingual dialog and watsonx.ai chat API access. Use when an Apache-2.0 IBM Granite model card and Hugging Face availability are important for review or deployment planning.

Getting Started

  1. Read the IBM Granite model documentation for Granite 4.0 and confirm the model identifier ibm/granite-4-h-small for watsonx.ai chat API usage.
  2. Review the official Hugging Face model card for Granite-4.0-H-Small
  3. including license
  4. supported capabilities
  5. evaluation notes
  6. training-data notes
  7. and limitations.
  8. Check IBM watsonx.ai foundation model documentation and pricing before production use
  9. including the documented input and output token rates for granite-4-h-small.
  10. If deploying outside watsonx.ai
  11. review the IBM Granite Hugging Face repository and any platform-specific deployment listing
  12. such as AWS Marketplace
  13. for infrastructure or endpoint billing notes.

Key Features

  • •Hybrid MoE architecture with 32B total parameters and 9B active parameters
  • •as stated in IBM’s Granite 4.0 launch announcement and documentation.
  • •Long-context instruction-following model with a documented 131
  • •072-token context window in IBM watsonx.ai foundation model documentation.
  • •Supports RAG
  • •agents
  • •tool use
  • •function calling
  • •FIM code
  • •multilingual dialog
  • •summarization
  • •and code-related tasks according to IBM documentation and the official Hugging Face model card.
  • •Official watsonx.ai chat API usage is documented with model_id ibm/granite-4-h-small.
  • •Official Hugging Face model card lists an October 2
  • •2025 release date and Apache-2.0 license.

Capabilities

  • •text generation
  • •instruction following
  • •retrieval-augmented generation
  • •agent workflows
  • •tool use
  • •function calling
  • •code generation
  • •fill-in-the-middle code
  • •summarization
  • •multilingual dialog

Last updated Jul 9, 2026

Granite 4.0 H Small · Discover · Kaino