Granite 4.2 30B
Granite 4.2 30B is the largest model in IBM's August 2026 Granite 4.2 release, a dense transformer under Apache 2.0. It reasons in think tags by default, with non-thinking and low-effort modes, and supports OpenAI-style function calling. The native context is 128K tokens, extendable to 512K.
Specifications
- Context window
- 128Kabout 98,000 words, or 200 pages
- Max output
- Not published
- Input price
- $0.16per 1M tokens
- Output price
- $0.65per 1M tokens
- Cached input
- $0.04per 1M tokens
- Weights
- OpenApache-2.0
- Parameters
- 29.3Bdense
- Input types
- Text
- Output types
- Text
- Reasoning mode
- Yes
- Tool calling
- Yes
- Released
- Aug 25, 2026Current
DeepInfra serverless price (131,072 context). Native context is 128K; IBM says it can be extended to 512K.
What it costs in practice
| Workload | Monthly cost at list price |
|---|---|
| Support chatbot30,000 conversations a month, 2,000 input and 400 output tokens each | $17 |
| Document summaries2,000 documents a month, 12,000 input and 800 output tokens each | $4.88 |
| Coding agent1,000 tasks a month, 60,000 input and 6,000 output tokens each | $14 |
Standard rates, no caching or batch discounts. Run your own numbers.
Strengths
- Apache 2.0 license
- Three reasoning modes, including low-effort
- Tested in 12 languages
Trade-offs to weigh
- Text-only input
- Dense 29B weights need about 59 GB at BF16
Where it fits
IBM positions Granite 4.2 30B for:
Where you can use it
- Hugging Face
- Ollama
- DeepInfra
Run it yourself
Released under Apache-2.0. Approximate memory needed just to load the weights:
| Precision | Memory for weights |
|---|---|
| 16-bit (BF16/FP16) | 59 GB |
| 8-bit (FP8/INT8) | 29 GB |
| 4-bit (Q4_K_M, about 4.85 bits) | 18 GB |
Add memory for the KV cache (it grows with context length and concurrent users) and 1-2 GB of runtime overhead. Size it precisely in the VRAM calculator.
Alternatives to consider
| Model | Tier | Context | Input / output per 1M | |
|---|---|---|---|---|
| Mistral Small 4Mistral AI | Efficient | 256K | $0.15 / $0.60 | Compare |
| Gemma 4 31BGoogle | Efficient | 256K | $0.20 / $0.40 | Compare |
| GLM-5.3-FlashZ.ai (Zhipu AI) | Efficient | 1M | $0.15 / $0.50 | Compare |
| Ministral 3 14BMistral AI | Efficient | 256K | $0.20 / $0.20 | Compare |
Frequently asked questions
How much does Granite 4.2 30B cost?
Granite 4.2 30B lists at $0.16 per million input tokens and $0.65 per million output tokens, with cached input at $0.04 (checked October 10, 2026). DeepInfra serverless price (131,072 context). Native context is 128K; IBM says it can be extended to 512K. Confirm current rates on the provider's pricing page before you commit.
What is the context window of Granite 4.2 30B?
Granite 4.2 30B accepts up to 131,072 tokens of context (about 98,000 words, or 200 pages). Long prompts cost more and can be slower, so most apps send only what the task needs.
Is Granite 4.2 30B open source?
Granite 4.2 30B is an open-weight model released under the Apache-2.0 license. You can download the weights from Hugging Face (ibm-granite/granite-4.2-30b) and run or fine-tune them yourself, within the license terms. Open weights isn't the same as fully open source: training data and code may not be published.
Can I run Granite 4.2 30B locally?
Yes, if you have enough memory. The weights alone need about 18 GB at 4-bit quantization and about 59 GB at 16-bit, plus room for the KV cache. Use the VRAM calculator on LLModel.com to check a specific GPU or Mac.
What is Granite 4.2 30B best for?
IBM positions Granite 4.2 30B for enterprise, self-hosting, agents, data extraction. It accepts text as input, has a reasoning (thinking) mode and supports tool calling. Test it on a sample of your own tasks before you choose it for production.
Sources
- huggingface.co/ibm-granite/granite-4.2-30b
- deepinfra.com/ibm-granite/granite-4.2-30b
- huggingface.co/ibm-granite/granite-4.2-30b/raw/main/config.json
- huggingface.co/api/models/ibm-granite/granite-4.2-30b
- www.ibm.com/granite
- ollama.com/library
Facts checked against these pages on October 10, 2026. Spot an error? Tell us.
Still choosing a model?
Send us your use case and constraints. We'll come back with a shortlist and an evaluation plan you can run on your own data.
- Free and vendor-neutral: no model provider pays for placement.
- Your details go only to partners you agree to.