Granite 4.1 3B Base
Granite 4.1 3B Base is a pretrained dense decoder model. It provides a starting point for specialized training and supports fill-in-the-middle code completion. It has not received the instruction alignment of the matching chat release.
Model design and training
The card links this checkpoint to the underlying Granite 4.0 Micro weights. It describes a multi-stage pretraining recipe of roughly 15 trillion tokens and additional long-context training. The architecture uses grouped-query attention and rotary position embeddings.Granite 4.1 3B Base model card
Practical use
Practical guidance. Choose this base checkpoint for domain adaptation or code-completion research. It does not supply the behavior of an instruction-tuned assistant. Keep base-model evaluations separate from chat evaluations, and record the tokenizer and prompt format used for each run.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-4.1-3b-base |
| Context window | 131,072 tokens Published configuration. The card separately describes a 512K context extension. Granite 4.1 3B Base model card |
| License | Apache 2.0Granite 4.1 3B Base model card |
| Parameters | 3BGranite 4.1 3B Base model card |
Related models
- Granite 4.2 3B
IBM · Granite 4.2
A 3B dense text model derived from Granite 4.1 Base with full, low-effort and direct-answer thinking modes.
Launch date Downloadable - Granite 4.1 3B
IBM · Granite 4.1
An instruction-tuned 3B text model for retrieval-based answers, extraction, coding and tool use.
Launch date API · Downloadable · Provider product