Today, we’re announcing the general availability of Meta’s Llama 3.2 vision models (90B and 11B) on the Oracle Cloud Infrastructure (OCI) Generative AI service. This release brings cutting-edge multimodal AI capabilities to OCI customers, enabling advanced image and text understanding in a single model. Launched by Meta on September 25, 2024, these models are competitive with leading foundation models in visual understanding and image recognition tasks.

Key highlights of this release include the following examples:

  • Both models support visual understanding and image recognition tasks
  • Both models offer a 128k context length
  • Llama 3.2 90B Vision includes the text-based capabilities of the previous Llama 3.1 70B model
  • Llama 3.2 11B Vision provides robust multimodal capabilities in a more compact form
  • Both models are available for dedicated hosting, with Llama 3.2 90B also offered for on-demand inferencing
  • Llama 3.1 70B remains available for on-demand, dedicated hosting, and fine-tuning

These models are available in the Chicago, London, and São Paulo regions. Use is restricted in the European Union. According to Meta’s Llama 3.2 Acceptable Use Policy, restrictions exist for individuals and companies in the European Union regarding the use of Llama 3.2 multimodal models. For details, review the full policy.

New features and improvements

The general availability release offers the following features and improvements:

  • Multimodal support: Text and image inputs with text output
  • Image understanding relevant for enterprise use cases, such as charts and graphs
  • Advanced image captioning and visual grounding capabilities
  • Multilingual support for text-only queries in eight languages

With the release of these new models, you can now achieve new image-based use cases, such as summarizing financial results from charts and graphs and writing captions for images and figures. Choose the Llama 3.2 90B model for highly accurate, highly performant results, or use the Llama 3.2 11B model for a more cost-effective option.

Want to know more?

OCI customers can now utilize these powerful multimodal models without infrastructure management concerns. Access is available through chat interfaces, APIs, or dedicated endpoints. For integration details and pricing information, consult our Generative AI service documentation or contact your Oracle representative.

For more information, see the following resources: