Granite 4.1 8b
ibm-granite · released 2026-04-30 · apache-2.0 license
IBM Granite 4.1 8B is a dense, non-reasoning Apache-2.0 small model (April 2026) tuned for enterprise tool calling and instruction following. Artificial Analysis index 12; per-benchmark numbers are unpublished as text.
Key specs
| Type | Local open-weight |
|---|---|
| Parameters | 8.79B total |
| Architecture | granite |
| Context window | 131K tokens |
| Knowledge cutoff | — |
| Modalities | text |
| Recommended backends | — |
| Minimum viable rig | — |
Benchmark scores
| GPQA Diamond | — |
|---|---|
| SWE-bench Verified | — |
| AIME | — |
| MMLU-Pro | — |
| BFCL v3 (tool use) | — |
| Composite score | — |
| Community rating | No reviews yet |
VRAM & disk per quantization
| Quant | VRAM | Disk | RAM | Context |
|---|---|---|---|---|
| Q4_K_M | 6.6 GB | 5.1 GB | — | 131K |
Strengths & weaknesses
Strengths: Strong tool calling and instruction following for its size; Dense 8B matches prior Granite 4.0 32B MoE on several evals; Fast, low cost, Apache-2.0, cryptographically signed
Weaknesses: Low overall intelligence vs frontier (AA Index 12); Non-reasoning model; IBM's numeric benchmark tables are images, none sourceable as text