v2.2 · 2026-06-12
IBM Granite 4.1 8B
granite-4-1-8b
IBM Granite 4.1 8B is a 8.79B text chat model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.
PROVIDER
IBM
FAMILY
IBM Granite
MODEL TYPE
text-chat
PARAMETERS
8.79B
MODALITIES
text
ARCHITECTURE
Granite 4.1 dense decoder-only transformer with grouped-query attention, RoPE, SwiGLU, RMSNorm and shared input/output embeddings.
CONTEXT WINDOW
131,072 tokens
TRAINING TOKENS
15.0T
RELEASE DATE
2026-04-29
GGUF REPOSITORIES
ibm-granite/granite-4.1-8b-GGUF official
Official IBM GGUF repository generated through IBM's GGUF conversion pipeline, with Q4_K_M, Q5_K_M and Q8_0 file listings.
SETUP HINTS
OLLAMA
Run `ollama run ibm/granite4.1:8b`.LM STUDIO
Use the official `ibm-granite/granite-4.1-8b-GGUF` repository and choose Q4_K_M for broad local compatibility.
LLAMA.CPP
Use a standard text GGUF loader, for example `llama-server -hf ibm-granite/granite-4.1-8b-GGUF:Q4_K_M`.QUANTIZATION FILE SIZES (GGUF)
Q4_K_M
5.35 GB
Q5_K_M
6.25 GB
Q8_0
9.35 GB
Quantization note: Q4_K_M for most enterprise local deployments; Q5_K_M on 16GB+ VRAM or 24GB+ RAM/unified-memory systems.
RAM / VRAM ESTIMATES
MIN RAM
10 GB
COMFORTABLE RAM
24 GB
MIN VRAM
8 GB
COMFORTABLE VRAM
16 GB
These are conservative local-inference estimates, not official hardware requirements.
HARDWARE FIT
CPU only (no GPU)
yes
8 GB RAM
limited
16 GB RAM
limited
32 GB RAM
yes
8 GB VRAM
limited
12 GB VRAM
yes
24 GB VRAM
yes
Apple Silicon (unified memory)
yes
Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.
BEST FOR
·Enterprise local chat
·RAG
·Structured JSON generation
·Tool calling
·Coding assistance
·Classification and extraction
·Long-context summarization on sufficiently large memory systems
·Commercial deployments needing Apache-2.0 licensing
AVOID IF
·You need vision or multimodal input
·You only have 8GB RAM and want long-context use
·You need the smallest possible edge model
·You need more than the verified language set
CAVEATS
·Chosen over Granite 3.3 8B Instruct because Granite 4.1 8B has newer official IBM model card, official IBM GGUF repository, IBM GGUF pipeline support and Ollama packaging.
·The model is marketed as 8B, while Ollama/official GGUF metadata reports about 8.79B to 9B parameters; `parameterSizeB` uses the more specific local artifact metadata.
·Although the context window is verified as long-context, using very large context lengths materially increases KV-cache memory.
·RAM and VRAM estimates are derived from GGUF/Ollama package size plus conservative overhead; KV cache, context length, batch size, runtime and GPU offload choices can materially change memory needs.
FIELD EVIDENCE
canonicalModelCardUrlhuggingface.co/ibm-granite/granite-4.1-8b
architecturehuggingface.co/ibm-granite/granite-4.1-8b
parameterSizeBollama.com/ibm/granite4.1:8b
releaseDatehuggingface.co/ibm-granite/granite-4.1-8b
contextWindowTokenshuggingface.co/ibm-granite/granite-4.1-8b
trainingTokensollama.com/ibm/granite4.1:8b
supportsToolshuggingface.co/ibm-granite/granite-4.1-8b
q4FileSizeGbhuggingface.co/ibm-granite/granite-4.1-8b-GGUF
q5FileSizeGbhuggingface.co/ibm-granite/granite-4.1-8b-GGUF
q8FileSizeGbhuggingface.co/ibm-granite/granite-4.1-8b-GGUF
officialGgufPipelinegithub.com/IBM/gguf
LANGUAGES
ArabicChineseCzechDutchEnglishFrenchGermanItalianJapaneseKoreanPortugueseSpanish
CAPABILITIES
tool callingreasoning tunedcode tuned