Model release

Granite Switch 4.1 8B

Granite Switch 4.1 8B is an open-weight model release in the IBM Granite family from IBM (Granite Team)1 5, version Switch 4.1 (8B), released Oct 14, 2026. Its weights can be downloaded by the public1 2 3 under the Apache License 2.01.

Granite Switch 4.1 8B is a composed checkpoint from IBM that pairs the Granite 4.1 8B language model with two activated-LoRA adapters (a general decision adapter and a Guardian safety-risk detector) and a speech-to-text cascade, so chat, decision, safety-check, and transcription requests can be served from one endpoint. Adapters are switched on through control tokens in the chat template.1 5 4Fact reviewed Oct 18, 2026

Model hub: Model card (external site: huggingface.co)
Repository: Granite Switch repository (external site: github.com)
Release notes: Launch blog (external site: huggingface.co)
License: Apache License 2.0 (Granite Switch repository) (external site: github.com)

On AI on Hugging Face: ibm-granite/granite-4.1-8b, ibm-granite/granite-speech-5.0-470m-turboctc, ibm-granite/granite-switch-4.1-8b, ibm-granite/granitelib-guardian-r1.0, ibm-granite/granitelib-core-r1.0

Last reviewedOct 18, 2026Entry updated Oct 18, 2026Documented release Oct 14, 2026

Availability and license

Overall availability

Public

Weights are downloadable from Hugging Face; the repository is not gated. Apache 2.0 terms apply.1 2

Availability fact review: Oct 18, 2026

Availability is separate from permission: read the license before using or redistributing.

Apache License 2.0 (external site: apache.org)1
Applies to: Weights · SPDX Apache-2.0 · on the rubric’s OSI-approved software-license list · fact reviewed Oct 18, 2026
Plain-language guide to the Apache License 2.0
Apache License 2.0 (external site: github.com)7
Applies to: Code · SPDX Apache-2.0 · on the rubric’s OSI-approved software-license list · fact reviewed Oct 18, 2026
Plain-language guide to the Apache License 2.0

The model card metadata and summary list Apache 2.0 for the checkpoint. The Granite Switch composition and inference code is published under Apache 2.0 in its repository LICENSE file.1 7Fact reviewed Oct 18, 2026

Component reuse rights

A readable or downloadable component is not automatically reusable. These indicators concern recorded license evidence, not system certification.

weights: Reviewed qualifying license recorded — check scope and conditions
code: Reviewed qualifying license recorded — check scope and conditions
data: Unknown — no complete fact-level rights review
documentation: Unknown — no complete fact-level rights review

No complete system-rights review is recorded for this release.

Model-disclosure tier

Computed from the checklist below using USASI rubric v0.2. An editorial category, not a certification.

Open-weight

The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.

How tiers are computed

Public materials checklist

Items for a model under USASI rubric v0.2. Unknown means unassessed or insufficient evidence.

Public materials checklist for Granite Switch 4.1 8B
Item	Status	Notes and evidence
WeightsCan the general public download the model parameters for this release?	Public	Bfloat16 safetensors shards with the adapters embedded are published in the Hugging Face repository.1 2 3
Inference codeIs code for running the model published?	Public	The model card documents inference with Hugging Face Transformers and vLLM through the open Granite Switch package, and the repository ships the chat template with the adapter activation logic.1 6
Training codeIs the code used to train the model published?	Unknown	The composition script that rebuilds this checkpoint from the base model and adapters is public, but this review did not assess training code for the base model, adapters, or speech model.
Training-data informationDoes the information cover provenance, scope, acquisition, selection, labeling, processing, and where data or alternatives can be obtained? Access alone does not establish completeness.	Unknown	Not assessed.
Training-data accessCan the training data be obtained? This is independent of information completeness and reuse rights; original unshareable data need not be downloadable.	Unknown	Not assessed.
Complete training pipelineIs the complete base-training and preprocessing pipeline published, including configuration? Fine-tuning code or an inference SDK alone is insufficient.	Unknown	Not assessed.
Legacy data assessment (v0.1)Historical assessment combining download access and disclosure. Preserved for traceability; excluded from the v0.2 tier calculation. See the new separate assessments above.	Unknown	Not assessed.
Training recipeAre the training configuration and procedure documented in enough detail to follow?	Unknown	BUILD.md documents the composition (adapter ranks, alpha values, and target modules). The training procedures of the composed components were not assessed in this review.
Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only.	Partial	The model card reports results for the decision adapter on a public decision benchmark and describes the invocation recipe; no evaluation code or configurations were found for this release.1

What it is useful for

The model card describes chat responses, multiple-choice decisions returned as probabilities over option labels, detection of safety risks such as harm or jailbreaking, and English speech transcription on the same endpoint.1

Run and use notes

Documented facts only. No hardware or performance claims are made without a cited source and stated assumptions.

The model card lists Python 3.9 or later, PyTorch 2.11.0 or later, Transformers 5.16 to 5.17.0, and optionally vLLM 0.26.0 up to (not including) 0.28.0, and documents serving the checkpoint with the vLLM OpenAI-compatible server.1
The model card describes the base model as having 8 billion parameters and a context length of 131,072 tokens; config.json sets max_position_embeddings to 131,072. BUILD.md reports 8,776,400,896 parameters after composing the base model with the two adapters.1 3 4
The speech-to-text cascade is English-only and returns lowercase, unpunctuated transcripts; the card notes the speech model and the language model are not jointly trained, so a mistranscription can change the response. The checkpoint's config.json leaves asr_model_id unset, which per the Granite Switch audio documentation selects ibm-granite/granite-speech-5.0-470m-turboctc and can be overridden.1 3 8

Organization context

People: Hendrik Strobelt, Kate Soule, Luis Lastras

Provenance and derivatives

Composed by IBM, without joint training, from its own Granite 4.1 8B instruct model, the guardian-core aLoRA from IBM's Granite Guardian Library r1.0, the decision aLoRA from IBM's Granite Core Library r1.0, and IBM's Granite Speech 5.0 470M TurboCTC speech model. Granite 4.1 8B is fine-tuned from IBM's Granite 4.1 8B Base.1 4 9 10 11 12

Derived from: Granite 4.1 8B (external site: huggingface.co) — IBM base language model; its card names the Granite Team, IBM as developer.
Derived from: Granite Guardian Library r1.0 (guardian-core aLoRA) (external site: huggingface.co) — IBM adapter library with adapters for Granite 4.0 and 4.1 models.
Derived from: Granite Core Library r1.0 (decision aLoRA) (external site: huggingface.co) — IBM adapter library with adapters for Granite 4.0 and 4.1 models.
Derived from: Granite Speech 5.0 470M TurboCTC (external site: huggingface.co) — IBM English speech-recognition model (conformer CTC encoder); its card says it was trained by IBM on public English audio corpora.

Other releases in the IBM Granite family

Granite 4.2 30BModel-disclosure tier (USASI rubric v0.2): Open-weight
Granite 4.2 3BModel-disclosure tier (USASI rubric v0.2): Open-weight
Granite 4.2 8BModel-disclosure tier (USASI rubric v0.2): Open-weight

IBM Granite family overview

What this catalog does not know

Unknown means the sources reviewed for this record do not document it. It is not evidence that something does not exist.

Training code: unknown.
Training-data information: unknown.
Training-data access: unknown.
Complete training pipeline: unknown.
Legacy data assessment (v0.1): unknown.
Training recipe: unknown.

Have a primary source? How to report a correction.

U.S. eligibility

Project eligibility rests on documented governing or maintaining entities, not on contributors.

Eligible · basis: U.S. headquarters

Published by IBM's Granite organization on Hugging Face and announced on IBM's Granite blog. IBM's corporate address is in Armonk, New York, per its U.S. contact page. The base model, both adapters, and the speech model are IBM releases; no other developer's weights are included by default.1 5 9 12 13

Assessed Oct 18, 2026

Sources

1.
ibm-granite/granite-switch-4.1-8b model card (external site: huggingface.co)
IBM (Hugging Face) · Model card · published Oct 14, 2026 · accessed Oct 18, 2026 · evidence reviewed Oct 18, 2026
2.
Hugging Face API metadata for ibm-granite/granite-switch-4.1-8b (external site: huggingface.co)
Hugging Face · Other · accessed Oct 18, 2026
3.
ibm-granite/granite-switch-4.1-8b config.json (external site: huggingface.co)
IBM (Hugging Face) · Repository · accessed Oct 18, 2026
4.
ibm-granite/granite-switch-4.1-8b BUILD.md (external site: huggingface.co)
IBM (Hugging Face) · Documentation · accessed Oct 18, 2026
5.
Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint (external site: huggingface.co)
IBM Granite (Hugging Face blog) · Announcement · published Oct 15, 2026 · accessed Oct 18, 2026
6.
Granite Switch repository README (external site: github.com)
Generative Computing (GitHub) · Repository · accessed Oct 18, 2026
7.
granite-switch LICENSE (external site: github.com)
Generative Computing (GitHub) · License · accessed Oct 18, 2026
8.
Granite Switch audio documentation (docs/AUDIO.md) (external site: github.com)
Generative Computing (GitHub) · Documentation · accessed Oct 18, 2026
9.
ibm-granite/granite-4.1-8b model card (external site: huggingface.co)
IBM (Hugging Face) · Model card · published Apr 29, 2026 · accessed Oct 18, 2026
10.
ibm-granite/granitelib-guardian-r1.0 model card (external site: huggingface.co)
IBM (Hugging Face) · Model card · accessed Oct 18, 2026
11.
ibm-granite/granitelib-core-r1.0 model card (external site: huggingface.co)
IBM (Hugging Face) · Model card · accessed Oct 18, 2026
12.
ibm-granite/granite-speech-5.0-470m-turboctc model card (external site: huggingface.co)
IBM (Hugging Face) · Model card · published Aug 25, 2026 · accessed Oct 18, 2026
13.
IBM contact page (United States) (external site: ibm.com)
IBM · Official page · accessed Oct 18, 2026

Email a correction Report a correction Edit this entry

This listing is not an endorsement, a safety assessment, or a federal approval.

Cite this page

APA style

USASI. (2026, October 18). Granite Switch 4.1 8B. United States of America Superintelligence. https://unitedstatesofamericasuperintelligence.com/open/granite-switch-4-1-8b/

BibTeX

@misc{usasi_granite_switch_4_1_8b,
  author = {{USASI}},
  title = {Granite Switch 4.1 8B},
  year = {2026},
  month = oct,
  howpublished = {\url{https://unitedstatesofamericasuperintelligence.com/open/granite-switch-4-1-8b/}},
  note = {United States of America Superintelligence. Last updated 2026-10-18}
}

The date is when this page was last updated. Add the date you read it if your style needs one.

For maintainers: a “Listed in USASI” badge

Links to this page. It says the entry is listed, not endorsed or rated. About the badge

Markdown (for a README)

[![Listed in the USASI catalog](https://unitedstatesofamericasuperintelligence.com/badges/listed.svg)](https://unitedstatesofamericasuperintelligence.com/open/granite-switch-4-1-8b/)

HTML

<a href="https://unitedstatesofamericasuperintelligence.com/open/granite-switch-4-1-8b/"><img src="https://unitedstatesofamericasuperintelligence.com/badges/listed.svg" alt="Listed in the USASI catalog" height="20"></a>