A collection of open source and IBM foundation models are available for inferencing in IBM watsonx.ai. Find foundation models that best suit the needs of your generative AI application and your budget.
The foundation models that are available for inferencing from watsonx.ai are hosted in various ways:
- Foundation models provided with watsonx.ai
- IBM-curated foundation models that are deployed on multitenant hardware by IBM and are available for inferencing. You pay by tokens used. See Foundation models provided with watsonx.ai.
- Deploy on demand foundation models
- An instance of an IBM-curated foundation model that you deploy and that is dedicated for your inferencing use. Only colleagues who are granted access to the deployment can inference the foundation model. A dedicated deployment means faster and more responsive interactions without rate limits. You pay for hosting the foundation model by the hour. See Deploy on demand foundation models.
- Custom foundation models
- Foundation models curated by you that you import and deploy in watsonx.ai. The instance of the custom foundation model that you deploy is dedicated for your use. A dedicated deployment means faster and more responsive interactions. You pay for hosting the foundation model by the hour. See Custom foundation models.
- Prompt-tuned foundation models
- A subset of the available foundation models that can be customized for your needs by prompt tuning the model from the API or Tuning Studio. A prompt-tuned foundation model relies on the underlying IBM-deployed foundation model. You pay for the resources that you consume to tune the model. After the model is tuned, you pay by tokens used to inference the model. See Prompt-tuned foundation models.
If you want to deploy foundation models in your own data center, you can purchase watsonx.ai software. For more information, see Overview of IBM watsonx as a Service and IBM watsonx.governance software.
Deployment methods comparison
To help you choose the right deployment method, review the comparison table.
| Deployment type | Available from | Deployment mechanism | Hosting environment | Billing method | Deprecation policy |
|---|---|---|---|---|---|
| Foundation models provided with watsonx.ai | • Resource hub>Pay per token • Prompt Lab |
Curated and deployed by IBM | Multitenant hardware | By tokens used | Deprecated according to published lifecycle |
| Deploy on demand foundation models | • Resource hub>Pay by the hour • Prompt Lab |
Curated and deployed by IBM at your request | Dedicated hardware | By hour deployed | Your deployed model is not deprecated |
| Custom foundation models | • Prompt Lab | Curated and deployed by you | Dedicated hardware | By hour deployed | Not deprecated |
| Prompt-tuned foundation models | • Prompt Lab | Tuned and deployed by you | Multitenant hardware | • Training is billed by CUH • Inferencing is billed by tokens used |
Deprecated when the underlying model is deprecated unless you add the underlying model as a custom foundation model |
For details on how model pricing is calculated and monitored, see Billing details for generative AI assets.
Supported foundation models by deployment method
Various foundation models are available from watsonx.ai that you can either use immediately or that you can deploy on dedicated hardware for use by your organization.
Foundation models provided with watsonx.ai
A collection of open source and IBM foundation models are deployed in IBM watsonx.ai. You can prompt these foundation models in the Prompt Lab or programmatically.
The following provided foundation models are deployed by IBM for inferencing in watsonx.ai:
- granite-13b-instruct-v2
- granite-8b-japanese
- granite-20b-multilingual
- granite-3-2b-instruct
- granite-3-8b-instruct
- granite-3-2-8b-instruct-preview-rc
- granite-guardian-3-2b
- granite-guardian-3-8b
- granite-3b-code-instruct
- granite-8b-code-instruct
- granite-20b-code-instruct
- granite-34b-code-instruct
- granite-ttm-512-96-r2
- granite-ttm-1024-96-r2
- granite-ttm-1536-96-r2
- allam-1-13b-instruct
- codellama-34b-instruct
- elyza-japanese-llama-2-7b-instruct
- flan-t5-xl-3b
- flan-t5-xxl-11b
- flan-ul2-20b
- jais-13b-chat
- llama-3-3-70b-instruct
- llama-3-2-1b-instruct
- llama-3-2-3b-instruct
- llama-3-2-11b-vision-instruct
- llama-3-2-90b-vision-instruct
- llama-guard-3-11b-vision-instruct
- llama-3-1-8b-instruct
- llama-3-1-70b-instruct
- llama-3-405b-instruct
- llama-3-70b-instruct (London and Sydney regions only)
- llama-2-13b-chat
- mistral-large
- mistral-small-24b-instruct-2501
- mixtral-8x7b-instruct-v01
- pixtral-12b
To start inferencing a provided foundation model, complete these steps:
- From the main menu, select Resource hub.
- Click View all in the Pay per token section.
- Click a foundation model tile, and then click Open in Prompt Lab.
IBM foundation models provided with watsonx.ai
The following table lists the supported IBM foundation models that IBM provides for inferencing.
Use is measured in Resource Units (RU); each unit is equal to 1,000 tokens from the input and output of foundation model inferencing. For details on how model pricing is calculated and monitored, see Billing details for generative AI assets.
Some IBM foundation models are also available from third-party repositories, such as Hugging Face. IBM foundation models that you obtain from a third-party repository are not indemnified by IBM. Only IBM foundation models that you access from watsonx.ai are indemnified by IBM. For more information about contractual protections related to IBM indemnification, see the IBM Client Relationship Agreement and IBM watsonx.ai service description.
| Model name | API model_id | Input price (USD/1,000 tokens) |
Output price (USD/1,000 tokens) |
Context window (input + output tokens) |
More information |
|---|---|---|---|---|---|
| granite-13b-instruct-v2 | ibm/granite-13b-instruct-v2 |
$0.0006 | $0.0006 | 8,192 | • Model card • Website • Research paper Note: This foundation model can be prompt tuned.
|
| granite-8b-japanese | ibm/granite-8b-japanese |
$0.0006 | $0.0006 | 4,096 | • Model card • Website • Research paper |
| granite-20b-multilingual | ibm/granite-20b-multilingual |
$0.0006 | $0.0006 | 8,192 | • Model card • Website • Research paper |
| granite-3-2b-instruct | ibm/granite-3-2b-instruct |
$0.0001 | $0.0001 | 131,072 | • Model card • Website • Research paper |
| granite-3-8b-instruct | ibm/granite-3-8b-instruct |
$0.0002 | $0.0002 | 131,072 | • Model card • Website • Research paper |
| granite-3-2-8b-instruct-preview-rc |
ibm/granite-3-2-8b-instruct-preview-rc |
No cost during tech preview | No cost during tech preview | 131,072 | • Model card • Website • Research paper |
| granite-guardian-3-2b | ibm/granite-guardian-3-2b |
$0.0001 | $0.0001 | 131,072 | • Model card • Website |
| granite-guardian-3-8b | ibm/granite-guardian-3-8b |
$0.0002 | $0.0002 | 131,072 | • Model card • Website |
| granite-3b-code-instruct | ibm/granite-3b-code-instruct |
$0.0006 | $0.0006 | 128,000 | • Model card • Website • Research paper |
| granite-8b-code-instruct | ibm/granite-8b-code-instruct |
$0.0006 | $0.0006 | 128,000 | • Model card • Website • Research paper |
| granite-20b-code-instruct | ibm/granite-20b-code-instruct |
$0.0006 | $0.0006 | 8,192 | • Model card • Website • Research paper |
| granite-34b-code-instruct | ibm/granite-34b-code-instruct |
$0.0006 | $0.0006 | 8,192 | • Model card • Website • Research paper |
| Model name | API model_id | Input price (USD/1,000 data points) |
Output price (USD/1,000 data points) |
Context length Min data points |
More information |
|---|---|---|---|---|---|
| granite-ttm-512-96-r2 | ibm/granite-ttm-512-96-r2 |
$0.00013 | $0.00038 | 512 | • Model card • Website • Research paper |
| granite-ttm-1024-96-r2 | ibm/granite-ttm-1024-96-r2 |
$0.00013 | $0.00038 | 1,024 • Model card • Website • Research paper |
|
| granite-ttm-1536-96-r2 | ibm/granite-ttm-1536-96-r2 |
$0.00013 | $0.00038 | 1,536 | • Model card • Website • Research paper |
Third-party foundation models provided with watsonx.ai
The following table lists the supported third-party foundation models that are provided with watsonx.ai.
Use is measured in Resource Units (RU); each unit is equal to 1,000 tokens from the input and output of foundation model inferencing. For details on how model pricing is calculated and monitored, see Billing details for generative AI assets.
| Model name | API model_id | Provider | Input price (USD/1,000 tokens) |
Output price (USD/1,000 tokens) |
Context window (input + output tokens) |
More information |
|---|---|---|---|---|---|---|
| allam-1-13b-instruct | sdaia/allam-1-13b-instruct |
National Center for Artificial Intelligence and Saudi Authority for Data and Artificial Intelligence | $0.0018 | $0.0018 | 4,096 | • Model card |
| codellama-34b-instruct | codellama/codellama-34b-instruct-hf |
Code Llama | $0.0018 | $0.0018 | 16,384 | • Model card • Meta AI Blog |
| elyza-japanese-llama-2-7b-instruct | elyza/elyza-japanese-llama-2-7b-instruct |
ELYZA, Inc | $0.0018 | $0.0018 | 4,096 | • Model card • Blog on note.com |
| flan-t5-xl-3b | google/flan-t5-xl |
$0.0006 | $0.0006 | 4,096 | • Model card • Research paper Note: This foundation model can be prompt tuned.
|
|
| flan-t5-xxl-11b | google/flan-t5-xxl |
$0.0018 | $0.0018 | 4,096 | • Model card • Research paper |
|
| flan-ul2-20b | google/flan-ul2 |
$0.0050 | $0.0050 | 4,096 | • Model card • UL2 research paper • Flan research paper |
|
| jais-13b-chat | core42/jais-13b-chat |
Inception, Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), and Cerebras Systems | $0.0018 | $0.0018 | 2,048 | • Model card • Research paper |
| llama-3-3-70b-instruct | meta-llama/llama-3-3-70b-instruct |
Meta | $0.00071 | $0.00071 | 131,072 | • Model card • Meta AI blog |
| llama-3-2-1b-instruct | meta-llama/llama-3-2-1b-instruct |
Meta | $0.0001 | $0.0001 | 131,072 | • Model card • Meta AI blog • Research paper |
| llama-3-2-3b-instruct | meta-llama/llama-3-2-3b-instruct |
Meta | $0.00015 | $0.00015 | 131,072 | • Model card • Meta AI blog • Research paper |
| llama-3-2-11b-vision-instruct | meta-llama/llama-3-2-11b-vision-instruct |
Meta | $0.00035 | $0.00035 | 131,072 | • Model card • Meta AI blog • Research paper |
| llama-3-2-90b-vision-instruct | meta-llama/llama-3-2-90b-vision-instruct |
Meta | $0.0020 | $0.0020 | 131,072 | • Model card • Meta AI blog • Research paper |
| llama-guard-3-11b-vision | meta-llama/llama-guard-3-11b-vision |
Meta | $0.00035 | $0.00035 | 131,072 | • Model card • Meta AI blog • Research paper |
| llama-3-1-8b-instruct | meta-llama/llama-3-1-8b-instruct |
Meta | $0.0006 | $0.0006 | 131,072 | • Model card • Meta AI blog |
| llama-3-1-70b-instruct | meta-llama/llama-3-1-70b-instruct |
Meta | $0.0018 | $0.0018 | 131,072 | • Model card • Meta AI blog |
| llama-3-405b-instruct | meta-llama/llama-3-405b-instruct |
Meta | $0.0050 | $0.016 | 16,384 | • Model card • Meta AI blog |
| llama-3-70b-instruct | meta-llama/llama-3-70b-instruct |
Meta | $0.0018 | $0.0018 | 8,192 | • Model card • Meta AI blog |
| llama-2-13b-chat | meta-llama/llama-2-13b-chat |
Meta | $0.0006 | $0.0006 | 4,096 | • Model card • Research paper |
| mistral-large | mistralai/mistral-large |
Mistral AI | $0.003 | $0.01 | 32,768 | • Model card • Blog post for Mistral Large 2 |
| mistral-small-24b-instruct-2501 | mistralai/mistral-small-24b-instruct-2501 |
Mistral AI | $0.00035 | $0.00035 | 32,768 | • Model card • Blog post for Mistral Small 3 |
| mixtral-8x7b-instruct-v01 | mistralai/mixtral-8x7b-instruct-v01 |
Mistral AI | $0.0006 | $0.0006 | 32,768 | • Model card • Research paper |
| mt0-xxl-13b | bigscience/mt0-xxl |
BigScience | $0.0018 | $0.0018 | 4,096 | • Model card • Research paper |
| pixtral-12b | mistralai/pixtral-12 |
Mistral AI | $0.00035 | $0.00035 | 128,000 | • Model card • Blog post for Pixtral 12B |
- For more information about the supported foundation models that IBM provides for embedding and reranking text, see Supported encoder models.
- For a list of which models are provided in each regional data center, see Regional availability of foundation model.
- For information about pricing and rate limiting, see watsonx.ai Runtime plans.
Custom foundation models
In addition to working with foundation models that are curated by IBM, you can upload and deploy your own foundation models. After the custom models are deployed and registered with watsonx.ai, you can create prompts that inference the custom models from the Prompt Lab and from the watsonx.ai API.
To learn more about how to upload, register, and deploy a custom foundation model, see Deploying a custom foundation model.
Deploy on demand foundation models
Choose a foundation model from a set of IBM-curated models to deploy for the exclusive use of your organization.
The following deploy on demand foundation models are available for you to deploy for inferencing in watsonx.ai:
- granite-3-8b-base
- granite-7b-lab
- granite-13b-chat-v2
- granite-13b-instruct-v2
- granite-20b-code-base-schema-linking
- granite-20b-code-base-sql-gen
- deepseek-r1-distill-llama-8b
- deepseek-r1-distill-llama-70b
- flan-t5-xl-3b
- flan-t5-xxl-11b
- flan-ul2-20b
- llama-2-13b-chat
- llama-2-70b-chat
- llama-3-8b-instruct
- llama-3-70b-instruct
- llama-3-1-8b
- llama-3-1-8b-instruct
- llama-3-1-70b-instruct
- llama-3-3-70b-instruct
- llama-3-3-70b-instruct-hf
- mistral-large-instruct-2407
- mistral-large-instruct-2411
- mistral-nemo-instruct-2407
- mixtral-8x7b-base
- mixtral-8x7b-instruct-v01
- mt0-xxl-13b
To deploy a deploy on demand foundation model, complete these steps:
- From the main menu, select Resource hub.
- Click View all in the Pay by the hour section.
- Click a foundation model tile, and then click Deploy.
For more information, see Deploying foundation models on-demand.
Deploy on demand foundation models from IBM
The following table lists the IBM foundation models that are available for you to deploy on demand.
Some IBM foundation models are also available from third-party repositories, such as Hugging Face. IBM foundation models that you obtain from a third-party repository are not indemnified by IBM. Only IBM foundation models that you access from watsonx.ai are indemnified by IBM. For more information about contractual protections related to IBM indemnification, see the IBM Client Relationship Agreement and IBM watsonx.ai service description.
| Model name | Price per hour in USD | Model hosting category | Context window (input + output tokens) |
|---|---|---|---|
| granite-7b-lab | $5.22 | Small | 4,096 |
| granite-13b-chat-v2 | $5.22 | Small | 8,192 |
| granite-13b-instruct-v2 | $5.22 | Small | 8,192 |
| granite-20b-code-base-schema-linking | $5.22 | Small | 8,192 |
| granite-20b-code-base-sql-gen | $5.22 | Small | 8,192 |
| granite-3-8b-base | $5.22 | Small | 4,096 |
Deploy on demand foundation models from third-parties
The following table lists the third-party foundation models that are available for you to deploy on demand.
| Model name | Provider | Price per hour in USD | Model hosting category | Context window (input + output tokens) |
|---|---|---|---|---|
| deepseek-r1-distill-llama-8b | DeepSeek AI | $5.22 | Small | 131,072 |
| deepseek-r1-distill-llama-70b | DeepSeek AI | $20.85 | Large | 131,072 |
| flan-t5-xl-3b | $5.22 | Small | 4,096 | |
| flan-t5-xxl-11b | $5.22 | Small | 4,096 | |
| flan-ul2-20b | $5.22 | Small | 4,096 | |
| llama-2-13b-chat | Meta | $5.22 | Small | 4,096 |
| llama-2-70b-chat | Meta | $20.85 | Large | 4,096 |
| llama-3-8b-instruct | Meta | $5.22 | Small | 8,192 |
| llama-3-70b-instruct | Meta | $20.85 | Large | 8,192 |
| llama-3-1-8b | Meta | $5.22 | Small | 131,072 |
| llama-3-1-8b-instruct | Meta | $5.22 | Small | 131,072 |
| llama-3-1-70b-instruct | Meta | $20.85 | Large | 131,072 |
| llama-3-3-70b-instruct | Meta | $10.40 | Medium | 131,072 |
| llama-3-3-70b-instruct-hf | Meta | $20.85 | Large | 131,072 |
| mixtral-8x7b-base | Mistral AI | $10.40 | Medium | 32,768 |
| mixtral-8x7b-instruct-v01 | Mistral AI | $10.40 | Medium | 32,768 |
| mistral-large-instruct-2407 | Mistral AI | $55.15 (See note.) | Large | 131,072 |
| mistral-large-instruct-2411 | Mistral AI | $55.15 (See note.) | Large | 131,072 |
| mistral-nemo-instruct-2407 | Mistral AI | $5.22 | Small | 131,072 |
| mt0-xxl-13b | BigScience | $5.22 | Small | 4,096 |
Hosting: $20.85 + Access: $34.30 = Total: $55.15 USD per hour
Prompt-tuned foundation models
You can customize the following foundation models by prompt tuning them in watsonx.ai:
For more information, see Tuning Studio.
Learn more
- Foundation model details
- Billing details for generative AI assets
- Regional availability of foundation model
Parent topic: Developing generative AI solutions