Multimodal foundation model

One AI model that can work with several kinds of information, such as reading text while also looking at an image.

Where it creates value

Best suited to work like this.

Reviewing documents with figures, interpreting images with notes, multimodal search, and assistant experiences that combine text, speech, or vision.

What it does

The intelligent task.

Cross-modal understanding, visual question answering, image or text generation, transcription, and combined reasoning across modalities.

Typical value

Combines text, images, tables, audio, or other evidence in one workflow, reducing handoffs between separate tools and analyses.

Where it performs well: Combines information that would otherwise require separate tools and manual reconciliation.

Value maturity · editorial assessment Demonstrated

Practical value is proven in multimodal pilots and selected production uses, but repeatable scaling patterns are still less mature than for text-only language models.

Emerging
Demonstrated
Scaling
Established

Back

TEC-0002 · How it works in practice

Multimodal foundation model

Read the card type together with the use case. A component can be central in one solution and enabling in another.

01 · Inputs

Text, images, audio, video, diagrams, tables, and task instructions.

02 · Intelligent task

Cross-modal understanding, visual question answering, image or text generation, transcription, and combined reasoning across modalities.

03 · Outputs

Text, labels, extracted fields, generated media, explanations, or structured observations.

Common application patterns

Reviewing documents with figures
interpreting images with notes
multimodal search
and assistant experiences that combine text
speech
or vision

Enablers

What must be in place.

Representative multimodal data, modality-specific quality checks, model access, evaluation by modality, and controls for sensitive media.

Where it struggles

Limitations and drawbacks.

Performance can differ sharply by modality; small visual details, numerical plots, or uncommon scientific images may be misread.

Human role

People remain accountable.

Subject-matter experts review interpretations and confirm that important visual or scientific detail was not missed.

Connected use cases

Examples from the current use-case library.

Accelerate image reviewAutomate preclinical image analysisPreclinical · Pilot Reduce manual reviewAutomate batch record reviewQuality · Concept

Continue exploring

The card is the simple front door. The website can hold the full system behind it.

Use the technology record to connect applications, processes, enablers, risks, controls, evidence, and implementation experience without placing all of it on the printable card.

Browse technologies