Foundation model lifecycle
To help you discover and use the latest and best foundation models, the list of foundation models that are available for prompting in watsonx.ai is updated regularly.
Foundation models that are built by IBM are continuously updated and improved. As new versions of IBM foundation models are introduced, older versions of these models are deprecated and eventually removed as they progress through the lifecycle over time.
Similarly, as newer and more effective third-party foundation models are released, older models progress through the same lifecycle in watsonx.ai so that the latest versions of a model family can be featured.
Modifications to IBM foundation models
IBM foundation models are periodically modified by IBM to improve the foundation model performance or security. A modification is a model refresh that might include new capabilities or fixes, but does not meet IBM's criteria to warrant a version update.
After a modification to an IBM foundation model is introduced, the modified version of the model automatically replaces the earlier version of the model when you upgrade your deployment.
Foundation model lifecycle stages
New foundation models from IBM and third-party providers are released frequently. To make sure you have access to the latest models, watsonx.ai manages a model's lifecycle from release to withdrawal from the platform.
Based on the lifecycle phase a model is in, foundation models in watsonx.ai can be categorized as follows:
- Legacy foundation model
- A foundation model moves to the legacy state when a newer version of a foundation model is released and available to install in the cluster. For example, when the third generation of an IBM model family is introduced, the second generation models become legacy models.
- Deprecated foundation model
- When a foundation model is deprecated in a software release version, you cannot install the deprecated model in a cluster from that software version moving forward. The model is scheduled to be withdrawn in a future release version.
- Withdrawn foundation model
- When a foundation model is withdrawn in a software release version, you cannot install the model in a cluster from that software version moving forward.
The following table compares the model lifecycle stages based on common criteria:
| Criteria | Legacy | Deprecated | Withdrawn |
|---|---|---|---|
| Availability | • Full access to the model with the same support as the latest models for inferencing, tuning, and more. • Recommended to switch to the latest models. |
• If the model was installed in a previous release before being deprecated, you can continue to access and inference the model. | • If the model was installed in a previous release before being withdrawn, you can continue to access and inference the model. • After the model is removed from the software and de-provisioned from the cluster, re-provision the model as a custom foundation model to continue using the model. For details, see Deploying custom foundation models. |
| Restrictions | Not applicable. | • Cannot perform a new installation of the model in the current software release. • Might be in a constricted state (inference only, no tuning, training, or deployment) if installed in a previous software release version. |
• Cannot perform a new installation of the model in the current software release. • No software support provided by IBM for future model updates. |
| Check model lifecycle stage | • For a list of legacy foundation models, see Supported foundation models. • The model can remain in legacy state for up to 1 year before being deprecated. |
• The model is highlighted with a warning icon  in the UI.• Warning message is generated by the API during inference. • The Foundation models by lifecycle state table shows the deprecated model, the release in which the model is deprecated, and a suitable alternative foundation model to consider as a replacement. |
• The model is highlighted with a warning icon  in the UI.• Warning message is generated by the API during inference. • The Foundation models by lifecycle state table shows the withdrawn model, the release in which thoe model is withdrawn, and a suitable alternative foundation model to consider as a replacement. |
Deprecated and withdrawn models
The following table lists the foundation models that are deprecated and scheduled for withdrawal from the watsonx platform:
| Foundation model name Foundation model ID API model ID |
Moved to legacy state in release: IBM watsonx experience version IBM Software Hub cluster version |
Deprecated in release: IBM watsonx experience version IBM Software Hub cluster version |
Withdrawn in release: IBM watsonx experience version IBM Software Hub cluster version |
Alternative foundation model |
|---|---|---|---|---|
| Name: pixtral-12b Model ID: mistralai-pixtral-12bAPI model ID: mistralai/pixtral-12b |
– | 2.3.0 5.3.0 |
To be determined | mistral-small-3-1-24b-instruct-2503 |
| Name: granite-vision-3-2-2b Model ID: ibm-granite-vision-3-2-2bAPI model ID: ibm/granite-vision-3-2-2b |
2.2.1 5.2.1 |
2.2.2 5.2.2 |
To be determined | granite-vision-3-3-2b |
| Name: mixtral-8x7b-instruct-v01 Model ID: mistralai-mixtral-8x7b-instruct-v01API model ID: mistralai/mixtral-8x7b-instruct-v01 |
– | 2.2.2 5.2.2 |
2.3.0 5.3.0 |
mistral-small-3-2-24b-instruct-2506 |
| Name: codestral-22b Model ID: codestral-22bAPI model ID: mistralai/codestral-22b |
2.2.2 5.2.2 |
2.2.2 5.2.2 |
2.3.0 5.3.0 |
codestral-2508 |
| Name: mistral-small-24b-instruct-2501 Model ID: mistral-small-24b-instruct-2501API model ID: mistralai/mistral-small-24b-instruct-2501 |
2.2.0 5.2.0 |
2.2.2 5.2.2 |
2.3.0 5.3.0 |
mistral-small-3-2-24b-instruct-2506 |
| Name: mistral-small-instruct Model ID: mistral-small-instructAPI model ID: mistralai/mistral-small-instruct |
2.1.2 5.1.2 |
2.2.2 5.2.2 |
2.3.0 5.3.0 |
mistral-small-3-2-24b-instruct-2506 |
| Name: mistral-large Model ID: mistral-largeAPI model ID: mistralai/mistral-large |
2.1.1 5.1.1 |
2.2.2 5.2.2 |
2.3.0 5.3.0 |
mistral-medium-2508 |
| Name: llama-4-scout-17b-16e-instruct Model ID: llama-4-scout-17b-16e-instructAPI model ID: meta-llama/llama-4-scout-17b-16e-instruct |
2.2.1 5.2.1 |
2.2.2 5.2.2 |
2.3.0 5.3.0 |
llama-4-scout-17b-16e-instruct-int4 |
| Name: llama-3-405b-instruct Model ID: llama-3-405b-instructAPI model ID: meta-llama/llama-3-405b-instruct |
2.2.0 5.2.0 |
2.2.2 5.2.2 |
2.3.0 5.3.0 |
llama-4-maverick-17b-128e-instruct-fp8 |
| Name: llama-2-13b-chat Model ID: meta-llama-llama-2-13b-chatAPI model ID: meta-llama/llama-2-13b-chat |
– | 2.2.2 5.2.2 |
2.3.0 5.3.0 |
– |
| Name: jais-13b-chat Model ID: core42-jais-13b-chatAPI model ID: core42/jais-13b-chat |
– | 2.2.2 5.2.2 |
2.3.0 5.3.0 |
– |
| Name: granite-13b-instruct-v2 Model ID ibm-granite-13b-instruct-v2 API model ID: ibm/granite-13b-instruct-v2 |
– | 2.2.1 5.2.1 |
2.3.0 5.3.0 |
granite-4-h-small |
| Name: flan-t5-xl-3b Model ID: google-flan-t5-xl-3bAPI model ID: google/flan-t5-xl-3b |
– | 2.2.1 5.2.1 |
2.3.0 5.3.0 |
– |
What to do next
You must choose an alternative supported foundation model to use if any of the following saved resources submit input to a foundation model that is withdrawn:
- Prompt template asset
- Prompt session asset
- Notebook asset
For details about working with saved prompt assets, see Saving your work.