For India — every deal says whether, and when, we checked it
doublethink
  1. India
  2. IBM Granite 4.0 Micro: free Apache 2.0 model weights for local AI
MiniCPM

IBM Granite 4.0 Micro: free Apache 2.0 model weights for local AI

Free
  • No end date
  • on huggingface.co
Get it on huggingface.co →

Opens the official offer page.

The offer

Users in India can download IBM Granite 4.0 Micro, a 3-billion-parameter instruction model, free from the official IBM Granite listing and run it on compatible hardware. The model card and IBM documentation state Apache 2.0, allowing commercial and personal use subject to its licence and notices. This is a downloadable model, not free hosted inference: you supply the computer, storage and electricity, and any cloud compute or hosted API service costs extra.

Who can claim it

  • Available for users in India under the worldwide Apache 2.0 licence.
  • Compatible local hardware or separately paid compute is required.

How to claim it

  1. Open the official IBM Granite 4.0 Micro model listing.
  2. Read the model card and Apache 2.0 licence.
  3. Download the weights and tokenizer files, or load ibm-granite/granite-4.0-micro with the model card Transformers instructions.
  4. Run the model on your own compatible hardware and review its outputs.

Good to know

The downloadable model has no stated trial expiry or subscription renewal. Hardware, electricity and any separately chosen hosted service remain your responsibility.

Preserve licence and attribution notices and mark changes when redistributing. The model is supplied as is and may produce inaccurate, biased or unsafe output. No free cloud credits, hosted API, IBM watsonx subscription or training-code bundle is included.

Source: the official offer page, huggingface.co/ibm-granite/granite-4.0-micro

Did it work for you?

Your vote is how a dead offer comes down: three or more "Didn't" in two days, outnumbering the "Worked" votes, and it is taken off the site.

Did it work?

Some links pay us a commission. It costs you nothing extra and never decides what is listed. Full disclosure.