Granite 3.0 8B
Granite 3.0 8B is a language model with 8B parameters from IBM. It is currently free at 1 provider with up to — tokens of context. Its weights are open, so it can also run locally.
- Developer
- IBM
- Parameters
- 8B
- Type
- Text
- License
- Open weights
Where to use Granite 3.0 8B for free
Each row is a separate free endpoint with its own limits.
| Provider | Context | Limit | Card | |
|---|---|---|---|---|
NVIDIA Build ibm/granite-3.0-8b-instruct |
— | Free serverless inference for development; limits and model availability may change. | No | Open |
Quick start
curl https://integrate.api.nvidia.com/v1/chat/completions \
-H "Authorization: Bearer $NVIDIA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "ibm/granite-3.0-8b-instruct",
"messages": [{"role": "user", "content": "Hello!"}]
}'from openai import OpenAI
client = OpenAI(
base_url="https://integrate.api.nvidia.com/v1",
api_key="YOUR_NVIDIA_API_KEY",
)
reply = client.chat.completions.create(
model="ibm/granite-3.0-8b-instruct",
messages=[{"role": "user", "content": "Hello!"}],
)
print(reply.choices[0].message.content)import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://integrate.api.nvidia.com/v1",
apiKey: process.env.NVIDIA_API_KEY,
});
const reply = await client.chat.completions.create({
model: "ibm/granite-3.0-8b-instruct",
messages: [{ role: "user", content: "Hello!" }],
});
console.log(reply.choices[0].message.content);Example for NVIDIA Build. Any OpenAI-compatible client works: Cursor, Cline, OpenCode, Open WebUI.
Benchmarks
This model is not on the Artificial Analysis leaderboard yet. We do not publish our own estimates.
Similar free models
FAQ
Is Granite 3.0 8B free?
Yes. As of September 26, 2026, Granite 3.0 8B is free at NVIDIA Build. Each provider has its own limits — see the table above.
How do I call Granite 3.0 8B via API?
Use any OpenAI-compatible client with base URL https://integrate.api.nvidia.com/v1, model ID ibm/granite-3.0-8b-instruct, and a NVIDIA Build key.
Is Granite 3.0 8B open-weight?
Yes, the weights are published — you can download and run it locally, for example with Ollama or LM Studio.