Supported foundation models in watsonx.ai

You can work with third-party and IBM foundation models in IBM watsonx.ai.

How to choose a model

To review factors that can help you to choose a model, such as supported tasks and languages, see Choosing a model and Foundation model benchmarks.

Provided foundation models that are ready to use

You can deploy foundation models from a collection of models curated by IBM in watsonx.ai. You can prompt these foundation models in the Prompt Lab or programmatically.

You can choose to deploy the following types of models:

IBM foundation models
Third-party foundation models

All IBM foundation models in watsonx.ai are indemnified by IBM.

For information about the GPU requirements for the supported foundation models, see Foundation models in IBM watsonx.ai in the IBM Software Hub documentation.

IBM foundation models

You can inference the following supported foundation models that are provided by IBM. The foundation models must be deployed in your cluster by an administrator to be available for use. All IBM models are instruction-tuned.

Note:

A view of all foundation models is available on the Resource hub.

Latest IBM foundation models

The following table lists the latest IBM foundation models in watsonx.ai for inferencing.

Table 1. Latest IBM foundation models in watsonx.ai for inferencing
Model name	Context window (input + output tokens)	Supported tasks	More information
ibm-defense-4-0-micro	131,072	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • function calling
granite-docling-258M	8,192	• extraction • generation • retrieval-augmented generation	• Website
granite-4-h-tiny	131,072	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • code • extraction • translation • function calling	• Granite 4.0 website
granite-4-h-micro	131,072	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • code • extraction • translation • function calling • code	• Granite 4.0 website
granite-4-h-small	131,072	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • code • extraction • translation • function calling • code	• Granite 4.0 website
ibm-defense-3-3-8b-instruct	128,000	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • code • extraction • translation • function calling	• Website
granite-vision-3-3-2b	131,072	• question answering • generation
granite-3-3-8b-instruct	131,072	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • code • extraction • translation • function calling	• Website
granite-guardian-3-2-5b	128,000	• classification • extraction • generation • question answering • summarization	• Website

Table 2. IBM foundation models provided with watsonx.ai for forecasting future values
Model name	Context length Min data points	More information
granite-ttm-512-96-r2	512	• Model card • Website • Research paper
granite-ttm-1024-96-r2	1,024	• Model card • Website • Research paper
granite-ttm-1536-96-r2	1,536	• Model card • Website • Research paper

Legacy IBM foundation models

The following table lists the legacy IBM foundation models in watsonx.ai for inferencing. When a newer version of a foundation model is released, the existing foundation models are moved to the legacy state. You can still access these legacy models with the same level of support as the latest versions, but switching to the newer models is recommended.

Table 3. Legacy IBM foundation models in watsonx.ai for inferencing
Model name	Context window (input + output tokens)	Supported tasks	More information
granite-3-2-8b-instruct	131,072	• classification • extraction • generation • question answering • summarization • retrieval-augmented generation • code • extraction • translation • function calling	• Website • Research paper
granite-3-2b-instruct	4,096	• classification • extraction • function calling • generation • question answering • summarization	• Website • Research paper
granite-3-8b-instruct	4,096	• classification • extraction • function calling • generation • question answering • summarization	• Website • Research paper
granite-guardian-3-2b	8,192	• classification • extraction • generation • question answering • summarization	• Website
granite-guardian-3-8b	8,192	• classification • extraction • generation • question answering • summarization	• Website
granite-20b-code-base-schema-linking	8,192	• code	• Research paper
granite-20b-code-base-sql-gen	8,192	• code	• Research paper
granite-8b-code-instruct	128,000	• code • classification • extraction • generation • question answering • summarization	• Website • Research paper
granite-vision-3-2-2b	131,072	• question answering	• Website • Research paper
granite-13b-instruct-v2	8,192	• classification • extraction • generation • question answering • summarization	• Website • Research paper

Third-party foundation models

You can inference the following supported third-party foundation models. An administrator must deploy the foundation models in your cluster before you can use these models.

Note:

A view of all foundation models is available on the Resource hub.

Latest third-party foundation models

The following table lists the supported foundation models that are provided by third parties.

Table 4. Latest third-party foundation models supported in watsonx.ai
Model name	Provider	Context window (input + output tokens)	Supported tasks	More information
codestral-2508	Mistral AI	256,000	• code	•Blog post for Codestral 25.08 Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
devstral-medium-2507	Mistral AI	128,000	• code	• Blog post for Codestral 25.01 Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
mistral-medium-2508	Mistral AI	131,072	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Medium 3 Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
mistral-small-3-2-24b-instruct-2506	Mistral AI	131,072	• classification • extraction • generation • question answering • retrieval-augmented generation • summarization • function calling • code • translation	• Research paper Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
gpt-oss-20b	OpenAI	131,072	• classification • extraction • generation • question answering • retrieval-augmented generation • summarization • function calling • code • translation	• OpenAI blog
gpt-oss-120b	OpenAI	131,072	• classification • extraction • question answering • retrieval-augmented generation • summarization • function calling • code • translation	• OpenAI blog
voxtral-small-24b-2507	Mistral AI	32,000	• classification • extraction • generation • question answering • retrieval-augmented generation • summarization • translation • function calling • audio understanding
llama-4-maverick-17b-128e-instruct-fp8	Meta	131,072	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization • translation • function calling	• Meta AI blog
llama-4-maverick-17b-128e-instruct-int4	Meta	128,000	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization • translation • function calling	• Meta AI blog
llama-4-scout-17b-16e-instruct-int4	Meta	128,000	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization • translation • function calling	• Meta AI blog
llama-3-3-70b-instruct	Meta	131,072	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog • Meta AI docs
llama-3-2-11b-vision-instruct	Meta	131,072	• classification • code generation and conversion • extraction • function calling • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog • Research paper
llama-3-2-90b-vision-instruct	Meta	131,072	• classification • code generation and conversion • extraction • function calling • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog • Research paper
llama-guard-3-11b-vision	Meta	131,072	• classification • code generation and conversion • extraction • function calling • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog • Research paper
ministral-8b-instruct	Mistral AI	128,000	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation • question answering • function calling	• Blog post for Ministral 8b Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
mistral-large-instruct-2411	Mistral AI	131,072	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Large 2
pixtral-large-instruct-2411	Mistral AI	128,000	• classification • generation • retrieval-augmented generation • summarization	• Blog post for Pixtral Large Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
allam-1-13b-instruct	National Center for Artificial Intelligence and Saudi Authority for Data and Artificial Intelligence	4,096	• classification • extraction • generation • question answering • retrieval-augmented generation • summarization • translation	• Research paper

Legacy third-party foundation models

The following table lists the legacy third-party foundation models in watsonx.ai for inferencing. When a newer version of a foundation model is released, the existing foundation models are moved to the legacy state. You can still access these legacy models with the same level of support as the latest versions, but switching to the newer models is recommended.

Table 5. Legacy third-party foundation models supported in watsonx.ai
Model name	Provider	Context window (input + output tokens)	Supported tasks	More information
mistral-medium-2505	Mistral AI	131,072	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Medium 3
mistral-small-3-1-24b-instruct-2503	Mistral AI	131,072	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Small 3
codestral-2501	Mistral AI	256,000	• code	• Blog post for Codestral 25.01 Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
llama-3-2-1b-instruct	Meta	131,072	• classification • code generation and conversion • extraction • function calling • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog • Research paper
llama-3-2-3b-instruct	Meta	131,072	• classification • code generation and conversion • extraction • function calling • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog • Research paper
llama-3-1-8b-instruct	Meta	131,072	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog
llama-3-1-70b-instruct	Meta	131,072	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog
mistral-large	Mistral AI	32,768	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Large 2 Note: You must purchase Mistral AI with IBM separately before you are entitled to use this model.
pixtral-12b	Mistral AI	128,000	• classification • generation • retrieval-augmented generation • summarization	• Blog post for Pixtral 12B
mistral-small-24b-instruct-2501	Mistral AI	32,768	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Small 3
codestral-22b	Mistral AI	32,768	• code	• Blog post for Codestral
flan-t5-xl-3b	Google	4,096	• classification • extraction • generation • question answering • retrieval-augmented generation • summarization	• Research paper
jais-13b-chat	Inception, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), and Cerebras Systems	2,048	• classification • extraction • generation • question answering • retrieval-augmented generation • summarization • translation	• Research paper
llama-4-scout-17b-16e-instruct	Meta	131,072	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization • translation • function calling	• Meta AI blog
llama-2-13b-chat	Meta	4,096	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization	• Research paper
mistral-small-instruct	Mistral AI	32,768	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Blog post for Mistral Small
llama-3-405b-instruct	Meta	16,384	• classification • code • extraction • generation • question answering • retrieval-augmented generation • summarization	• Meta AI blog
mixtral-8x7b-instruct-v01	Mistral AI	32,768	• classification • code • extraction • generation • retrieval-augmented generation • summarization • translation	• Research paper

Learn more

IBM foundation models
Third-party foundation models
For more information about the foundation models that IBM provides for embedding and reranking text, see Supported encoder models.