opendunes

Granite 4.0 Micro

by Ibm GraniteReleased 20 October 2025

Granite 4.0 Micro is a small text model from IBM, tuned to use the tools and apps you connect to it. It reads about 131 pages at once, so you can pass a long thread or a pile of documents along with your instructions and get text back. Use it for small, quick steps in an automated workflow.

TakesText
GivesText

Reads about 130 pages · writes up to 110 pages

Parameters you can send

Standard OpenAI fields, grouped by what they control. Anything not listed is accepted and ignored.

Shaping the reply

How long, how varied, where to stop.

max_tokenstemperaturetop_ptop_kmin_pstopseedfrequency_penaltypresence_penaltyrepetition_penaltylogit_biaslogprobstop_logprobs

Structured output

Get JSON you can parse instead of prose.

response_format

Questions about Granite 4.0 Micro

  • Is Granite 4.0 Micro still on sale?

    No. Ibm Granite has retired it. Other Ibm Granite models are on sale.

  • What is Granite 4.0 Micro best for?

    Granite 4.0 Micro is a small text model from IBM, tuned to use the tools and apps you connect to it. It reads about 131 pages at once, so you can pass a long thread or a pile of documents along with your instructions and get text back. Use it for small, quick steps in an automated workflow.

  • Does Granite 4.0 Micro work in Arabic, French or Darja?

    IBM lists 12 languages: English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch and Chinese. Darja is not named.

  • Can it read PDFs, images or a whole book?

    It reads text. Up to 131K tokens in one request, about 130 pages. It writes up to 88 000 words at once.

  • Is Granite 4.0 Micro open source?

    Yes, its weights are public. Open weights on Hugging Face

  • Who makes Granite 4.0 Micro, and when was it released?

    Ibm Granite. Released 20 October 2025.

Other Ibm Granite models, priced in dinars

See all Ibm Granite models
Granite 4.2 8BIbm Granite · Aug 31, 2026

Granite 4.2 8B is an IBM text model that thinks before it answers, for maths, code, multilingual chat and automated tasks that need several moves. For each request you choose full thinking, light thinking or none, and it can use the tools you connect to it. It reads about 131 pages at once and works in 12 languages, including Arabic, Japanese and Korean.

Thinks firstUses your tools130 pages at once
≈0,062 DAper message
View model