Granite 3.3 8B
by IBM · granite family
8B
parameters
text-generation code-generation reasoning math tool-use summarization
Granite 3.3 8B is IBM's enterprise-focused language model built on a dense transformer decoder architecture with GQA, RoPE, and SwiGLU. It supports a 128K context window and features structured reasoning through dedicated thinking and response tags. The model delivers strong results on AlpacaEval-2.0 and Arena-Hard benchmarks, with particular strengths in mathematics, coding, and instruction adherence. Released under the Apache 2.0 license, it is well-suited for enterprise RAG applications and business use cases.
Quick Start with Ollama
ollama run 8b-instruct-q8_0 | Creator | IBM |
| Parameters | 8B |
| Architecture | transformer-decoder |
| Context | 125K tokens |
| Released | Apr 16, 2025 |
| License | Apache 2.0 |
| Ollama | granite3.3 |
Quantization Options
| Format | File Size | VRAM Required | Quality | Ollama Tag |
|---|---|---|---|---|
| Q4_K_M | 4.1 GB | 6 GB | | 8b-instruct-q4_K_M |
| Q8_0 rec | 7.2 GB | 10 GB | | 8b-instruct-q8_0 |
| F16 | 15.2 GB | 18 GB | | 8b-instruct-fp16 |
Compatible Hardware
Q8_0 requires 10 GB VRAM