Granite Models by IBM Research
About
The Granite family of large language models, developed by IBM, is tailored for enterprise applications. These open-source models are available under the Apache 2.0 license and highlight performance, safety, and cost-efficiency. The latest iteration, Granite 3.0, marks a significant upgrade, being trained on over 12 trillion tokens in 12 languages and 116 programming languages. Offering a variety of model sizes and specializations, including the safety-focused Granite Guardian and latency-optimized Mixture-of-Experts (MoE) models, they brim with versatility. These models can handle tasks such as summarization, text classification, and code generation. They're accessible across platforms like Hugging Face, IBM watsonx, and Google Cloud's Vertex AI. IBM ensures transparency and responsible AI utilization through comprehensive documentation and benchmarks, underscoring their commitment to open-source innovation 123.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
Use when the workload needs 8k context and 13B parameters.
Use when the workload needs 8k context and 20B parameters.
Use when the workload needs 4k context and 8B parameters.
Use when the workload needs 8k context and 13B parameters.
Use when the workload needs 8k context and 7B parameters.
Use when the workload needs 8k context and 13B parameters.
Use when the workload needs 8k context and 13B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Granite 3B | Use when the workload needs 3B parameters. | 2024-10 | 3B parameters | Current |
| Granite 7B | Use when the workload needs 7B parameters. | 2024-10 | 7B parameters | Current |
| Granite 34B | Use when the workload needs 34B parameters. | 2024-10 | 34B parameters | Current |
| Granite 13B Chat | Use when the workload needs 8k context and 13B parameters. | 2023-11 | 8k context13B parameters | Current |
| Granite 20B Multilingual | Use when the workload needs 8k context and 20B parameters. | 2023-11 | 8k context20B parameters | Current |
| Granite 8B Japanese | Use when the workload needs 4k context and 8B parameters. | 2023-11 | 4k context8B parameters | Current |
| Granite 13B | Use when the workload needs 8k context and 13B parameters. | 2023-11 | 8k context13B parameters | Current |
| Granite 7B Lab | Use when the workload needs 8k context and 7B parameters. | 2023-11 | 8k context7B parameters | Current |
| IBM Granite 13B Instruct v1 | Use when the workload needs 8k context and 13B parameters. | 2023-11 | 8k context13B parameters | Current |
| IBM Granite 13B Chat v1 | Use when the workload needs 8k context and 13B parameters. | 2023-11 | 8k context13B parameters | Current |
Release Timeline
2 release groupsSpecifications(10 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Granite 3B | 2024-10 | — | 3B |
| Granite 7B | 2024-10 | — | 7B |
| Granite 34B | 2024-10 | — | 34B |
| Granite 13B Chat | 2023-11 | 8k | 13B |
| Granite 20B Multilingual | 2023-11 | 8k | 20B |
| Granite 8B Japanese | 2023-11 | 4k | 8B |
| Granite 13B | 2023-11 | 8k | 13B |
| Granite 7B Lab | 2023-11 | 8k | 7B |
| IBM Granite 13B Instruct v1 | 2023-11 | 8k | 13B |
| IBM Granite 13B Chat v1 | 2023-11 | 8k | 13B |
Available From(1 provider)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| IBM Granite 13B Instruct v1 | IBM watsonx | $0.185 | $0.185 | Serverless |
| IBM Granite 13B Chat v1 | IBM watsonx | $0.185 | $0.185 | Serverless |
| Granite 20B Multilingual | IBM watsonx | $0.6 | $0.6 | Serverless |
| Granite 13B Chat | IBM watsonx | $0.6 | $0.6 | Serverless |
| Granite 13B | IBM watsonx | $0.6 | $0.6 | Serverless |
| Granite 8B Japanese | IBM watsonx | $0.6 | $0.6 | Serverless |
| Granite 7B Lab | IBM watsonx | $0.6 | $0.6 | Serverless |
Frequently Asked Questions
- What is Granite used for?
- Granite is used for coding. The family description and listed model capabilities point to those workloads as the best fit.
- How does Granite compare to Granite 4?
- Granite by IBM Research is strongest where you need coding, while Granite 4 by IBM Research is the closest related family to check for audio. Granite has 10 listed variants and reaches up to 8k context, while Granite 4 reaches up to 131k context, so compare the specs and pricing tables before choosing a production model.
- Which Granite model should I use?
- For the lowest listed input price, start with IBM Granite 13B Instruct v1 through IBM watsonx at $0.185/1M input tokens. For the most capable/latest local choice, evaluate IBM Granite 13B Instruct v1 with 8k context.
Related Model Families
Models(10)
Granite 3B
Granite 7B
Granite 34B
Granite 13B Chat
Granite 20B Multilingual
Granite 8B Japanese
Granite 13B
Granite 7B Lab
IBM Granite 13B Instruct v1
IBM Granite 13B Chat v1