BastionGPT vs Nabla: AI Medical Scribe Comparison for 2026

Pricing you can see, budget, and act on today

Nabla does not publish pricing. Its website routes buyers to a demo request, and quotes come through a sales conversation. Reported pricing for its Pro tier has been approximately $119 per user per month (Axios and multiple 2025-2026 industry reviews), with enterprise contracts priced by custom quote, and 2026 roundups report Nabla now sells primarily through demo and contract.

BastionGPT publishes its prices and has held them steady for over two years. Professional starts at $20 per user per month. Professional Plus, which adds the full document analysis suite, image and spreadsheet reading, and manual model selection, costs $45. Teams that need SSO and deployment inside their own enterprise cloud also get a published number: Ultra at $75 per user per month. Annual billing includes a free month, and every plan carries a 45-day money-back guarantee.

The math for a five-clinician practice: at Nabla's reported $119 rate, a year costs $7,140 ($119 x 5 x 12). BastionGPT Professional Plus runs $2,700 ($45 x 5 x 12), and Professional runs $1,200. That is a difference of $4,440 to $5,940 per year, and with BastionGPT you know the number before you ever talk to anyone.

‍

A full AI assistant that keeps working after the note is signed

Nabla is a documentation platform. It transcribes the encounter, generates the note, and suggests ICD-10 codes. Per Glass Health's 2026 analysis, Nabla does not answer clinical questions, search literature, or generate content the clinician did not verbalize during the visit; the assessment and plan reflect what was said in the room.

BastionGPT includes an unlimited general AI assistant on every plan. Clinicians use it to draft referral letters, insurance appeal letters, treatment plans, and patient education materials; review documentation for incorrect names, pronouns, vague language, and suspected under- or over-coding; and analyze lab summaries and spreadsheets. A Saved Prompts library with folder organization and team sharing ships on every plan, so a practice can standardize documentation by sharing one vetted prompt set and inserting any prompt with a slash command. BastionGPT received 99.3% sentence-level accuracy on clinical documentation tasks from a peer-reviewed study by Baylor College of Medicine and Texas Children's Hospital researchers.

The distinction is simple: Nabla ends when the note is signed. BastionGPT keeps working on the prior authorization, the appeal, the records summary, and the parent-friendly explanation of the diagnosis.

‍

Document analysis Nabla does not offer

On Professional Plus and Ultra plans, BastionGPT reads 10+ documents at once, up to 1,000+ pages of combined context. Supported formats include PDF, Word, Excel, CSV, PowerPoint, HTML, plain text, and JPG or PNG images, including scanned PDFs, charts, and graphs. A clinician can upload a full patient history, a psychological evaluation, and records from multiple providers, then query, summarize, and cross-reference all of it in one conversation. The Professional plan covers a single PDF, Word, or text document up to 30 pages.

Nabla's product is built around the encounter: live ambient audio and at-cursor dictation. It does not offer a general document upload and analysis workflow, and it does not accept spreadsheets, records, or images for review. For practices that handle records requests, independent evaluations, legal referrals, or insurance documentation, that is an entire category of work Nabla leaves on your desk.

‍

Your notes stay yours for more than two weeks

Nabla retains encounter data for 14 days by default, a window its help center describes as time to review, edit, and export notes to the EHR. The default is configurable at the organization level, and audio is not stored at all. That minimal-retention posture is a genuine security strength for hospital deployments, and a real constraint for an individual clinician: if a note was not exported within the window, it is gone.

BastionGPT wipes chats and transcripts after 30 days by default, and each user controls the window: retain records longer when a case is still open, or delete them sooner, including immediate deletion of specific items or all data. You get the same zero-training guarantee with retention that fits how a practice actually works.

One more difference worth knowing: Nabla's help center states that optional user feedback, which may include de-identified audio with PHI replaced by a beep, is stored indefinitely and used both to train Nabla and for troubleshooting. BastionGPT never uses customer inputs, uploads, or audio to train AI models, with no exceptions to document.

‍

AI models you can see and choose

BastionGPT names its engines: licensed and HIPAA-compliant versions of GPT-5.5, Claude, and Gemini 3 Pro. Every plan selects the best model for each task automatically, and Professional Plus adds manual model selection plus early access to new models as they ship. Model quality drives note quality, and naming the models lets you verify exactly what is reading your records.

Nabla runs on a model that is not disclosed publicly, so you cannot benchmark it against alternatives, or switch to a stronger engine for a complex task.

‍

A scribe built for real-world sessions

BastionGPT recognizes up to 10 speakers per session with speaker diarization, and its AI automatically labels each speaker's name and role in the transcript: clinician, patient, parent, interpreter. Sessions run up to 240 minutes, and BastionGPT transcribes both live audio and uploaded recordings, so a dictation captured in the car or a recorded team meeting produces the same structured output as a live visit.

Nabla supports multi-speaker encounters but publishes no speaker count, and its workflow centers on live capture: ambient visits up to 3 hours, and dictated notes capped at 20 to 30 minutes per its help center. Its developer API accepts uploaded audio files, but that capability is for software builders, not a workflow inside the clinician app.

‍

A signed BAA on every plan, plus the compliance depth US practices need

BastionGPT enters a signed Business Associate Agreement with every customer on every plan, including the free trial, and supports HIPAA, 42 CFR Part 2, FERPA, PIPEDA, PHIPA, and Australian APP requirements. Data is encrypted with AES-256 at rest and TLS in transit, runs on HITRUST CSF certified and SOC 2 Type II attested infrastructure, and is independently penetration tested at least annually. In three years serving more than 10,000 healthcare organizations, BastionGPT has had zero breaches of protected health information.

Nabla maintains a strong public-facing enterprise security posture: HIPAA and GDPR compliance, SOC 2 Type 2 and ISO 27001 certifications, Texas RAMP Level 2, and a BAA incorporated into its Terms of Service for clients. Third-party reviews report that the free tier does not include a BAA, which matters because using a scribe on patient encounters without a BAA is itself a HIPAA compliance gap. Nabla also does not list 42 CFR Part 2 or FERPA support, which substance-use treatment programs and school-based providers should note before piloting.

‍

Human support and a published uptime guarantee

BastionGPT provides email, chat, phone, and video support with real humans on every plan, backed by a 99.90% uptime guarantee. Nabla's self-serve users rely on self-service support, with dedicated customer success reserved for enterprise contracts, and it publishes no uptime commitment outside those contracts.

‍