Model library / ibm-granite
Granite Switch 4.1 30B is a modular LLM combining a 30-billion base, decision and guardian adapters, and an English speech-to-text encoder.
Parameters in total
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
Granite Switch 4.1 30B is a composable language model released by IBM Granite. It bundles a 30-billion-parameter base model with two aLoRA adapters, a general-purpose decision adapter and a guardian-core safety adapter, and an embedded English speech-to-text encoder. The model is intended to serve chat, decision-making, and transcription tasks from a single endpoint, with adapters activated via control tokens.
In practice the model occupies roughly 30 billion parameters and stores weights in bfloat16 format. It supports a context window of up to 131,072 tokens, allowing long-form interactions. The licence is Apache-2.0, permitting commercial use. No mixture-of-experts layers are present. It can be run with HuggingFace Transformers or vLLM backends.
Specification
| Repository | ibm-granite/granite-switch-4.1-30b |
| Publisher | ibm-granite |
| Published | 14 October 2026 |
| Architecture | GraniteSwitchForCausalLM |
| Model type | granite_switch |
| Parameters | 29,989,814,272 |
| Layers | 66 |
| Hidden size | 4,096 |
| Attention heads | 32, 8 key/value heads |
| Context length | 131,072 tokens |
| Vocabulary | 100,355 tokens |
| Weight format | bfloat16 |
| Licence | apache-2.0 |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 55.9 GiB | Full precision as released |
| fp8 | 27.9 GiB | 8-bit, near-lossless on most models |
| int4 | 14.0 GiB | 4-bit, smallest footprint |
The cache costs 264 KiB per token at 16-bit, so the full 131,072-token context of one request needs about 33.0 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 4 | 20.5 GiB | about 81,575 tokens |
| RTX 5090 32GB | 4 | 49.3 GiB | the full 131,072 tokens |
| L40S 48GB | 2 | 25.5 GiB | about 101,435 tokens |
| A100 80GB | 1 | 13.6 GiB | about 54,169 tokens |
| H100 80GB | 1 | 13.6 GiB | about 54,169 tokens |
| RTX PRO 6000 96GB | 1 | 28.0 GiB | about 111,365 tokens |
| H200 141GB | 1 | 68.5 GiB | the full 131,072 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve ibm-granite/granite-switch-4.1-30b \ --max-model-len 77824 \ --gpu-memory-utilization 0.90
Questions
The weights take about 56 GiB at the released precision. That fits on one A100 80GB, which leaves roughly 14 GiB for the key-value cache.
The repository declares apache-2.0. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 131,072 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email hello@aiqre.com with the model name and roughly what volume you expect.
Also in the library
Granite Switch 4.1 8B is a modular 8.8 billion-parameter language model that combines a base LLM, decision and safety adapters, a…
Granite-4.2-30B is a 29.3 billion-parameter decoder-only language model from IBM, designed for reasoning tasks with built-in chai…
Granite-4.2-8B is an 8.8 billion-parameter decoder-only language model from IBM, designed for reasoning-intensive tasks.
Granite-4.2-3B is a 3.7 billion-parameter decoder-only language model with built-in chain-of-thought reasoning, released by IBM u…
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.