Granite 4 Models by IBM Research
Last refreshed 2026-09-18. Next refresh: weekly.
Details
About
Granite 4 is IBM’s Granite 4.0 language-model family, built around a hybrid Mamba-2/transformer design with Mixture-of-Experts in selected sizes. IBM documents the line for enterprise workloads such as RAG, multi-agent workflows, and edge or low-latency use, with sizes from small hybrid MoE workhorse models (for example Granite-4.0-H-Small) down to micro and sub-billion dense options when Mamba-2 support is limited. The docs also cover tool calling, RAG-style document grounding, fill-in-the-middle coding, and structured JSON output.
Decision facts
- Best fit
- audiocodingagent workflows
- Capability starting point
- Granite 4.0 H Micro with 131k context
- Lowest tracked input
- Granite 4.0 H Micro · $0.017/1M · OpenRouter
- Closest related family
- Granite 4.1
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs audio and 1B parameters.
Use when the workload needs 128k context and 3B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Granite 4.0 1B Speech | Use when the workload needs audio and 1B parameters. | 2026-03 | audio1B parameters | Current |
| Granite 4.0 | Use when provider availability and model metadata match the workload. | 2025-11 | — | Current |
| Granite 4.0 1B | Use when the workload needs 128k context. | 2025-10 | 128k context | Current |
| Granite 4.0 350M | Use when the workload needs 128k context. | 2025-10 | 128k context | Current |
| Granite 4.0 H 1B | Use when the workload needs 128k context. | 2025-10 | 128k context | Current |
| Granite 4.0 H 350M | Use when the workload needs 128k context. | 2025-10 | 128k context | Current |
| Granite 4.0 H Small | Use when the workload needs 128k context. | 2025-10 | 128k context | Current |
| Granite 4.0 H Tiny | Use when the workload needs 128k context. | 2025-10 | 128k context | Current |
| Granite 4.0 Micro | Use when the workload needs 128k context and 3B parameters. | 2025-10 | 128k context3B parameters | Current |
| Granite 4.0 H Micro | Use when the workload needs 131k context. | 2025-04 | 131k context | Current |
Release Timeline
4 release groupsSpecifications(10 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Granite 4.0 1B Speech | 2026-03 | — | 1B |
| Granite 4.0 | 2025-11 | — | — |
| Granite 4.0 1B | 2025-10 | 128k | ~1B |
| Granite 4.0 350M | 2025-10 | 128k | ~400M |
| Granite 4.0 H 1B | 2025-10 | 128k | ~1.5B |
| Granite 4.0 H 350M | 2025-10 | 128k | ~350M |
| Granite 4.0 H Small | 2025-10 | 128k | 32B total / 9B active |
| Granite 4.0 H Tiny | 2025-10 | 128k | 7B total / 1B active |
| Granite 4.0 Micro | 2025-10 | 128k | 3B |
| Granite 4.0 H Micro | 2025-04 | 131k | — |
Available From(2 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Granite 4.0 H Micro | OpenRouter | $0.017 | $0.112 | Serverless |
Related Model Families
Models(10)
Granite 4.0 1B Speech
Granite 4.0
Granite 4.0 1B
Granite 4.0 350M
Granite 4.0 H 1B
Granite 4.0 H 350M
Granite 4.0 H Small
Granite 4.0 H Tiny
Granite 4.0 Micro
Granite 4.0 H Micro