Scaling Abridge for Nurses at the Bedside with OpenAI’s Frontier Models

Nursing documentation is fundamentally different from narrative clinical notes: structured and distributed across flowsheets that span assessments, interventions, vitals, intake and output, pain, mobility, safety, and more. Each individual entry has to reflect what was communicated or observed at the bedside, then map to the right place in the health system’s EHR. Reasoning from a natural bedside conversation to a structured set of clinically supported documentation entries becomes a distinct AI challenge.

As OpenAI’s models continue to advance, Abridge evaluates how each generation can best support the demands of real clinical workflows. In the case of documentation for bedside nurses, GPT-5.4 has proven especially effective for structured flowsheet generation, where outputs need to be discrete, supported by the conversation, and aligned to the nurse’s workflow. This reflects Abridge’s broader approach: applying the right model for the right clinical purpose, with careful attention to quality, latency, reliability, and workflow fit.

In deployments of Abridge’s latest nursing flowsheet architecture with GPT-5.4, Abridge drafted approximately 30–40% more flowsheet fields from the bedside conversation versus the previous generation for nurse review.¹ By more comprehensively capturing the care that nurses deliver at the bedside, Abridge helps reduce the burden of documentation while creating more space for patient care, supporting continuity and coordination across care teams.

Designing improved architecture for nursing documentation

Capturing structured data from bedside conversations requires a foundation built specifically for nursing documentation, where the output format, validation, and workflow expectations mandate more focus on structure than narrative.

The model needs to deconstruct the nurse-patient conversation into clinically meaningful domains, evaluate what is supported by the conversation, and map that information into the appropriate flowsheet structure for each health system.

“Across every shift and any number of patients, nurses capture a large amount of clinical detail that belongs in hundreds of possible flowsheets rows. The challenge is determining which details are supported by the bedside conversation, resolving where they belong in the record, and presenting them clearly for nurse validation.”

– Peng Xu, Staff Machine Learning Scientist, Abridge

Frontier models are essential to this work, and their impact depends on how they are integrated into clinical workflows. For nursing, Abridge pairs OpenAI’s model capabilities with a purpose-built architecture designed for structured outputs. That architecture brings together a clinical ontology, EHR-specific flowsheet schemas, and partner-specific evaluation working in harmony together to reflect the realities of nursing documentation across different health system environments.

Pairing GPT-5.4 with nursing-specific architecture to drive completeness and relevance

Earlier models helped establish the foundation for structured nursing documentation, showing that information from a bedside conversation could meaningfully support flowsheet drafts. With GPT-5.4, we saw a step-change in the model’s abilities across complex nurse-patient conversations to support more complete flowsheet generation.

GPT-5.4 Scorecard vs. GPT-5.2

‍

“What changed was our ability to pair frontier models with an architecture purpose-built for nursing documentation. The model better identifies clinical detail from natural bedside conversation, and the architecture transforms that into structured scales, assessments, and flowsheet-ready outputs.”

– Noah Weissman, GenAI Software Engineer, Abridge

In 2026, Abridge began rolling out its next-generation nursing flowsheet architecture alongside GPT-5.4, designed to improve how the system identifies, structures, and drafts nursing documentation from bedside conversations. The comparison above holds the new architecture constant to isolate the model-level gain seen with GPT 5.4. In production, Abridge’s latest nursing flowsheet architecture with GPT-5.4 drafted approximately 30–40% more flowsheet fields from the bedside conversation compared to the previous generation for nurse review.¹

That architecture helps the model work across the full context of a bedside conversation, producing more complete documentation across assessments and scales while supporting deeper customization for the documentation standards and clinical language that vary across organizations and care settings. Together, OpenAI’s model capabilities and Abridge’s architecture helped turn complex bedside conversations into structured, nurse-reviewed outputs that fit real-world flowsheet workflows.

“Abridge continues to demonstrate how advanced AI can meaningfully improve clinical workflows when paired with deep healthcare expertise and rigorous evaluation. Their latest work using GPT-5.4 within a nursing-specific architecture shows the potential for AI to reduce documentation burden while helping care teams more comprehensively capture the work happening at the bedside. We’re excited to support Abridge as they continue expanding what’s possible with AI in healthcare.”

– Barret Zoph, GM of B2B, OpenAI

Since nursing documentation becomes part of the medical record, performance cannot be evaluated by reviewing the final output alone. Each generated entry has to be assessed against what the conversation supports, then reviewed by a nurse before it is filed to the chart.

Internal evaluation includes annotated datasets reviewed by registered nurses, as well as testing with partner health systems before and during deployment. Health systems can also use Abridge’s reporting infrastructure to review generated documentation. All Abridge-generated documentation is presented to the nurse for review, without any AI-generated documentation entered into the medical record before clinician confirmation. The final documentation reflects the care delivered and the nurse’s judgment about what belongs in the record.

What this means for nurses and health systems

Structured capture enhancements are about more than volume. They enable the system to recognize more of the assessments, interventions, observations, and education nurses are already communicating at the bedside, and present that information back in a format that nurses can quickly review and approve.

For a nurse on the floor, the difference shows up in how the shift feels: less time reconstructing what happened after the fact, fewer flowsheet rows entered manually, and more documentation that reflects the enormity of the care delivered.

The value extends beyond: the conversation at the bedside, the education a patient needs to go home safely, and the coordination across the care team that keeps the day moving.

“The best adoption happens when nurses see that the technology fits the way they already care for patients. When the right flowsheet rows are drafted from a natural bedside conversation, nurses don’t have to stop and translate their care into documentation from scratch.”

— Alyssa Stauffacher, RN, Director Clinical Success – Nursing, Abridge

Looking ahead: GPT-5.5 and what’s next for nursing AI

Our focus is on where stronger models meaningfully enhance structured documentation quality, domain coverage, and performance in complex clinical scenarios. For nursing documentation, GPT-5.4 proved especially effective for structured flowsheet generation, where the task is to produce reliable, reviewable outputs that fit naturally into the nurse’s workflow. As Abridge looks ahead to models like GPT-5.5, the goal is the same: apply the right model for the right purpose at the right time, balancing quality, latency, reliability, and clinical workflow fit.

Stronger structured documentation is just one step toward a broader platform for team-based care. Nursing work does not happen in isolation but remains difficult to surface in today’s systems, even though it shapes outcomes and patient experience. Abridge is building toward clinical AI that makes that work more visible, connects it across nurses, physicians, and the broader care team, and creates more space for clinicians to stay present with the patients at the center of the system.

To talk with our team about bringing AI-powered nursing documentation to your health system, contact us. Interested in an AI engineering role at Abridge? We’re hiring.