Suki AI Pricing Breakdown 2026: The Token-Cap Problem Costing Internal Medicine Groups Thousands

Clinical Update — June 2026: This guide has been revised to reflect the CMS CY2026 Physician Fee Schedule final rule adjustments to G2211 valuation, updated USCDI v3 Clinical Notes conformance requirements effective Q2 2026, and new FHIR R4 Composition resource handling guidance from ONC's USCDI v3 specifications. Token-cost modeling tables have been recalculated using observed 2026 inference pricing from major LLM API providers. If you read an earlier version, start at Section 2 for the revised overage math.

TL;DR for the CMIO: Every 2026 Suki AI pricing breakdown—including Freed's—compares sticker prices ($299–$399/mo) without examining what happens after the session hits the token cap. For a 20–50 provider internal medicine group managing complex chronic conditions (HFrEF + CKD-4, multi-problem CCM), "Advanced Reasoning" token overages on MDM narratives can silently inflate per-clinician costs by 15–30% above the listed tier. Worse, token-capped note generation often truncates the longitudinal relationship language required for G2211, leading to downcodes and add-on denials. This playbook is the first to model those hidden costs, map them to specific ICD-10 pairs and CPT workflows, and show how Scribing.io's problem-oriented FHIR-native architecture eliminates both the overage and the documentation gap.

• What Every 2026 Pricing Breakdown Misses: Token Caps and Complex Chronic Care

• Suki AI Pricing Tiers Deconstructed: Where the Overages Hide

• Scribing.io Clinical Logic: High-Stakes HFrEF + CKD-4 Case Walkthrough

• Technical Reference: ICD-10 Documentation Standards for I50.22 and N18.4

• The G2211 + CCM Revenue Trap: Why Token-Capped Scribes Fail MDM

• FHIR R4 Architecture: Why Whole-Note Re-Generation Is the Real Cost Driver

• Side-by-Side: Suki AI vs. Freed vs. Scribing.io for Internal Medicine Groups

• Action Plan: How Your CMIO Should Evaluate AI Scribe TCO in 2026

What Every 2026 Pricing Breakdown Misses: Token Caps and Complex Chronic Care

The competitor landscape in 2026—Freed's comparison page included—frames the Suki AI pricing conversation around monthly subscription tiers: $299/month for Compose, $399/month for Assistant, and enterprise custom pricing above $400/clinician/month. Freed positions itself as the cost-efficient alternative at $39–$119/month. Scribing.io exists because neither analysis addresses the variable that actually determines total cost of ownership for an internal medicine group: token consumption on complex clinical reasoning.

Strip away the marketing and examine what both Freed and Suki's published materials fail to disclose: token caps on specialized clinical reasoning are not uniform across visit types. A straightforward acute visit—URI, 99213—consumes a fraction of the inference tokens that a multi-problem chronic care visit requires. When an AI scribe must perform cross-problem risk synthesis (loop diuretic titration in HFrEF constrained by CKD-4 GFR decline), medication reconciliation justification across 12+ active medications with renal dosing adjustments, longitudinal relationship attestation language required for the G2211 complexity add-on per AMA CPT guidance, and data review documentation spanning labs, imaging, prior authorizations, and specialist notes—the session's token demand can exceed the "standard reasoning" allocation and silently trigger what Suki's architecture categorizes as "Advanced Reasoning," a premium inference tier.

For context on how documentation density varies by discipline and why a single token model cannot serve all specialties, see how we handle Psychiatry DAP notes (short, affect-heavy, minimal cross-problem synthesis) versus Family Medicine workflows (moderate complexity, high volume, broad ICD-10 surface area). Internal medicine sits at the apex of token demand because of sustained multi-problem MDM narratives.

Current clinical benchmarks from JAMA Internal Medicine documentation studies indicate that complex chronic care MDM narratives for patients with 3+ active high-risk conditions average 900–1,400 words of structured documentation. A standard token allocation designed for 400–600 word encounter notes cannot accommodate this without escalation.

The Math for a 30-Provider Internal Medicine Group

These figures do not include the downstream revenue loss from truncated MDM narratives that fail G2211 or CCM (99487/99489 per CMS fee schedule) audits—a cost quantified in the G2211 section below.

Freed's comparison page mentions Suki's "$379/clinician/month net revenue gain" but does not interrogate what happens when the AI's reasoning engine hits its ceiling on exactly the visits that generate the highest reimbursement. This is the gap we fill.

Suki AI Pricing Tiers Deconstructed: Where the Overages Hide

To conduct a meaningful Suki AI pricing breakdown in 2026, a CMIO must distinguish between list price, effective price, and total cost of ownership.

List Price (Published)

Hidden Variable Costs a CMIO Must Audit

1. Token-Cap Overages on Advanced Reasoning. Suki's tiered model requires scrutiny of token caps on specialized clinical reasoning; practices must identify if "Advanced Reasoning" for complex chronic care triggers additional "Premium Tier" billing. Suki allocates a standard inference budget per session. When the reasoning engine engages on tasks requiring multi-step clinical logic—cross-problem risk stratification, polypharmacy interaction analysis, synthesizing longitudinal relationship documentation—the session may escalate to a Premium Tier inference path. Practices must request from Suki:

• The exact token allocation per session at each pricing tier

• The threshold that triggers Advanced Reasoning escalation

• The per-token or per-session surcharge for Premium Tier inference

• Whether CCM and G2211-specific reasoning tasks are classified as standard or advanced

2. Whole-Note Re-Generation Costs. On EHRs still running FHIR R4 (which lacks a discrete "MDM rationale" resource element), Suki typically writes reasoning into Composition.section.text. Any late clinician edit—correcting a medication dose, adding a missed problem, updating a lab value—forces re-generation of the entire section. Each re-generation is a new inference call that counts against the token budget.

3. Implementation and Integration Costs. Unlike browser-extension tools, Suki's deep EHR integrations require IT involvement, Epic App Orchard provisioning, and often a 60–90 day deployment cycle. For a 30-provider group, current benchmarks indicate implementation costs of $15,000–$40,000 depending on EHR complexity.

Questions to Ask During Suki Contract Negotiation

1. "What is the per-session token ceiling for our internal medicine panel mix?"

2. "Can you provide a utilization report showing how many of our sessions would have triggered Advanced Reasoning based on our visit complexity distribution?"

3. "Is the $379 net revenue gain figure calculated before or after token overages and implementation costs?"

4. "How are CCM documentation sessions (99487/99489) billed against the token budget—separately or bundled with the face-to-face encounter?"

Scribing.io Clinical Logic: High-Stakes HFrEF + CKD-4 Case Walkthrough

This section demonstrates exactly how Scribing.io handles the visit type that breaks token-capped AI scribes.

The Scenario

A PCP in a one-party consent state sees a 72-year-old established patient with HFrEF (I50.22) and CKD Stage 4 (N18.4). Active medication list: 14 medications including sacubitril/valsartan, carvedilol, furosemide, spironolactone (with K+ monitoring constraints from CKD-4), dapagliflozin, and others. The visit involves medication reconciliation, lab review (BMP, BNP, CBC from 3 days prior), discussion of volume status, and shared decision-making about diuretic adjustment. The clinician intends to bill 99215 + G2211 and initiate CCM 99487 for the upcoming month.

What Happens with a Token-Capped Scribe

Estimated financial impact of this single visit:

• Premium token overage: $1.50–$2.50

• G2211 denial: ~$16.04 (2026 national average)

• 99215 → 99214 downcode: ~$55 revenue loss

• CCM 99487 denial: ~$93 revenue loss

• Total per-visit risk: $165.54–$166.54

Multiply across a panel of 200+ complex chronic patients per provider per year, and the exposure for a 30-provider group reaches six figures annually.

What Happens with Scribing.io: Step-by-Step Logic Breakdown

Step 1: On-Device Pre-Trim. Before any audio reaches the cloud, Scribing.io's on-device processing layer strips non-billable content—social pleasantries, repeated instructions, caregiver side-conversations, ambient noise artifacts. Only clinically relevant speech segments are tokenized. For this 22-minute encounter, pre-trim reduces the cloud-bound transcript from ~4,800 tokens to ~2,100 tokens. Token savings: 56%.

Step 2: Problem-Oriented Capture and ICD-10 Binding. Instead of generating a monolithic narrative, Scribing.io's intake engine parses the trimmed transcript into discrete problem threads. Each thread is bound to its ICD-10 code at capture time:

• Problem 1: HFrEF → I50.22. Captured content: volume status assessment, BNP trend, furosemide dose discussion, ejection fraction reference.

• Problem 2: CKD-4 → N18.4. Captured content: GFR trend (19 → 17 over 6 months), potassium 5.1, spironolactone hold discussion, nephrology referral consideration.

• Cross-problem thread: Diuretic-renal interaction, SGLT2i renal threshold (dapagliflozin continuation at eGFR 17 per KDIGO 2024 guidance).

Step 3: MDM Micro-Template Generation (Per Problem). Each problem thread generates a compact MDM micro-template of fewer than 180 words, mapped to USCDI v3 Clinical Notes data classes. This is where Scribing.io diverges fundamentally from token-capped architectures: instead of one large inference call that synthesizes all problems into a single narrative, the system makes targeted, small inference calls per problem. Each micro-template contains:

• Problem statement with ICD-10

• Data reviewed (labs, imaging, prior notes) with dates

• Risk assessment language per AMA 2021 E/M guidelines Table of Risk

• Management decision and rationale

• Medication changes with clinical justification

Token cost per micro-template: ~320 tokens. Two problems + cross-problem thread = ~960 tokens of inference. Compare to the monolithic approach: 1,800–2,400 tokens for the same clinical content, with the additional risk of exceeding the cap.

Step 4: MDM Gap Detector and G2211 Justification Prompt. Before the note is finalized, Scribing.io's MDM Gap Detector runs a rule-based audit (not an additional LLM call—zero extra tokens) against the generated micro-templates. For this visit, the Gap Detector checks:

1. G2211 Longitudinal Relationship Statement: Does the note contain an explicit statement that this clinician has an ongoing relationship with the patient for management of these conditions? The transcript mentions "we've been adjusting your heart failure medications together for the past three years." The Gap Detector flags this as present but not yet rendered in the MDM. It prompts the clinician: "Confirm G2211 relationship statement: 'Ongoing longitudinal management of HFrEF and CKD-4 since [date], with iterative medication optimization and renal function monitoring.' Accept/Edit?"

2. Risk Driver Attribution: Are both I50.22 and N18.4 explicitly linked to their risk drivers (drug therapy requiring intensive monitoring, decision regarding hospitalization)? The Gap Detector verifies each micro-template contains the required risk language.

3. Data Review Completeness: Are all reviewed data elements (BMP, BNP, CBC, prior echocardiogram, nephrology consult note) documented with dates? Missing items are flagged for clinician confirmation.

Step 5: Per-Problem FHIR R4 Composition Writes. Here is the architectural advantage that eliminates whole-note re-generation. Instead of writing all MDM content into a single Composition.section.text block, Scribing.io writes each problem's micro-template as a separate Composition.section with its own section.code (mapped to LOINC 51848-0 for Assessment) and section.entry references to the relevant Condition, MedicationStatement, and Observation resources.

When the clinician edits the furosemide dose from 40mg to 60mg after signing, Scribing.io regenerates only the HFrEF micro-template section (~320 tokens), not the entire note. The CKD-4 section, the G2211 justification section, and the CCM documentation section remain untouched. Each write includes a FHIR Provenance resource recording the agent (AI-assisted), timestamp, and reason for revision.

Step 6: CCM Documentation Separation. The CCM 99487 initiation documentation is generated as a standalone Composition resource, linked to but distinct from the office visit note. It includes the care plan summary, anticipated monthly coordination activities, patient consent documentation, and time-tracking framework—all required by CMS CCM billing requirements. This separation ensures the CCM claim is defensible independent of the E/M visit.

Result

• Total token spend: ~2,100 (pre-trim transcript) + ~960 (micro-template inference) + ~320 (G2211 justification) = ~3,380 tokens. No cap exceeded. No overage.

• G2211 add-on: Preserved. Explicit longitudinal relationship language present, clinician-confirmed.

• 99215 level: Supported. Per-problem risk attribution meets high-complexity MDM threshold.

• 99487 CCM: Standalone documentation ready for separate claim submission.

• Edit cost: Single micro-template re-gen (~320 tokens) vs. whole-note re-gen (~2,400 tokens).

Technical Reference: ICD-10 Documentation Standards for I50.22 and N18.4

The HFrEF + CKD-4 combination is one of the highest-risk ICD-10 pairings in internal medicine, and it is precisely the pairing most likely to trigger documentation deficiencies under token-constrained AI systems. Proper coding requires maximum specificity to prevent denials.

I50.22 - Chronic systolic (congestive) heart failure; N18.4 - Chronic kidney disease represents the foundational code pair for this patient. Each code demands specific documentation elements that Scribing.io enforces at capture time:

I50.22 — Chronic Systolic (Congestive) Heart Failure

• Specificity requirement: The note must document systolic (not diastolic, not unspecified) and chronic (not acute, not acute-on-chronic). Truncated AI narratives frequently default to I50.9 (unspecified) when the MDM section runs out of token budget before specifying type and acuity.

• Supporting documentation: Most recent ejection fraction with date, NYHA functional class, current GDMT (guideline-directed medical therapy) status per ACC/AHA heart failure guidelines.

• Scribing.io enforcement: The HFrEF micro-template requires EF value, NYHA class, and GDMT checklist as mandatory fields. If the transcript does not contain an EF reference, the system prompts: "EF not detected in encounter. Last documented EF: 30% (echo 2025-11-14). Carry forward? Y/N."

N18.4 — Chronic Kidney Disease, stage 4 (severe)

• Specificity requirement: Stage must be explicitly documented. CKD without stage defaults to N18.9 (unspecified), which does not support the risk stratification needed for 99215 MDM. The stage must correspond to the most recent eGFR per KDIGO staging criteria (Stage 4 = eGFR 15–29).

• Supporting documentation: Most recent eGFR with date, trend direction, etiology if known, current nephrology co-management status.

• Scribing.io enforcement: The CKD micro-template auto-populates stage from the most recent eGFR lab value via FHIR Observation query. If the eGFR has crossed a stage boundary since last visit (e.g., eGFR dropped from 31 to 17, crossing from Stage 3b to Stage 4), the system alerts: "CKD stage change detected: N18.3b → N18.4. Update problem list? Y/N." This prevents the common error of carrying forward a stale CKD stage code.

Cross-Code Documentation Requirements

When I50.22 and N18.4 coexist, the note must document the clinical interaction between the conditions—specifically, how heart failure management is constrained by renal function and vice versa. Per the AMA Table of Risk, drug therapy requiring intensive monitoring (RAAS inhibitors + diuretics in CKD-4) qualifies as high risk. Scribing.io's cross-problem thread captures this interaction explicitly, linking the two micro-templates with a shared risk statement.

The G2211 + CCM Revenue Trap: Why Token-Capped Scribes Fail MDM

G2211 (Visit complexity inherent to evaluation and management associated with medical care services that serve as the continuing focal point for all needed health care services) was introduced in CY2024 and has become a critical revenue component for internal medicine groups managing complex chronic populations. The 2026 national average reimbursement is approximately $16.04 per claim.

For a 30-provider group where each provider sees 12 G2211-eligible patients per week, the annual G2211 revenue at stake is:

30 providers × 12 visits/week × 48 weeks × $16.04 = $277,171/year

G2211 requires three documentation elements that token-capped scribes consistently fail to generate:

1. Longitudinal relationship statement: An explicit attestation that the billing clinician has an ongoing relationship with the patient for management of one or more chronic conditions. This is not a standard element in ambient documentation templates—it requires the scribe to synthesize information from prior visits, not just the current encounter.

2. Complexity drivers: Documentation that the conditions being managed create complexity beyond what is captured in the E/M level alone—typically the interaction between multiple chronic conditions.

3. Focal point language: A statement that this clinician serves as the continuing focal point for the patient's care coordination.

Token-capped systems face a structural problem: G2211 justification language is low clinical value but high billing value. It does not describe a new finding, a new medication, or a new assessment—it describes a relationship. AI systems optimizing for clinical content within a token budget will deprioritize this language every time, because it does not contribute to the clinical narrative. The result: G2211 denials on audit.

Scribing.io solves this with the MDM Gap Detector described in Step 4 above. The G2211 justification is a rule-based prompt, not an LLM inference call. It costs zero additional tokens. It fires automatically when the billing code 99215 is detected and G2211 is eligible. The clinician confirms or edits a pre-populated statement. Done.

Similarly, CCM 99487 requires documentation of at least 60 minutes of clinical staff time per calendar month for patients with two or more chronic conditions expected to last at least 12 months. When this documentation is entangled in the office visit note (as happens with monolithic note generators), payers routinely deny the CCM claim for insufficient standalone documentation. Scribing.io's separate CCM Composition resource eliminates this failure mode entirely.

FHIR R4 Architecture: Why Whole-Note Re-Generation Is the Real Cost Driver

Most internal medicine groups on Epic, Oracle Health, or athenahealth are running FHIR R4 interfaces for third-party application integration. FHIR R4's Composition resource is the standard vehicle for clinical notes. The problem: FHIR R4 does not define a discrete "MDM rationale" resource type. There is no MedicalDecisionMaking resource. There is no RiskAssessment resource with MDM-specific semantics.

Vendors—Suki included—therefore pack MDM reasoning into Composition.section.text, a narrative XHTML block. This block is atomic at the section level: you cannot update a single problem's assessment without touching the entire section. The consequences for token economics are severe:

Scribing.io achieves this granularity by writing each problem's MDM content as a separate Composition.section with distinct section.code values. Each section carries its own FHIR Provenance resource, creating a complete audit trail that satisfies both HIPAA audit requirements and payer documentation requests. When a section is updated, only that section's Provenance is regenerated—the rest of the note's provenance chain remains intact.

This architecture also enables USCDI v3 conformance. The ONC USCDI v3 Clinical Notes data class requires that clinical notes be exchangeable with sufficient granularity for care coordination. A single monolithic Composition.section.text blob does not meet this standard for multi-problem encounters. Scribing.io's per-problem sections, with entry references to discrete Condition, Observation, and MedicationStatement resources, achieve true USCDI v3 conformance.

Side-by-Side: Suki AI vs. Freed vs. Scribing.io for Internal Medicine Groups

Action Plan: How Your CMIO Should Evaluate AI Scribe TCO in 2026

Pricing breakdowns that stop at the monthly subscription line item are not evaluations—they are brochure summaries. Here is the audit framework your CMIO should run before signing or renewing any AI scribe contract:

Phase 1: Visit Complexity Profiling (Week 1)

1. Pull your CPT distribution. What percentage of your group's E/M claims are 99214 vs. 99215? What is your G2211 attach rate? What is your CCM (99487/99489) volume?

2. Identify your "token-heavy" visit types. Any visit with 3+ active chronic conditions, medication reconciliation of 10+ drugs, or cross-specialty data review will exceed standard token allocations.

3. Calculate your token-heavy visit frequency. For most internal medicine groups, this is 30–50% of encounters. If your group skews toward complex chronic care (geriatrics, cardiorenal, diabetes management), it may exceed 60%.

Phase 2: TCO Modeling (Week 2)

1. Request token utilization data from your current vendor. If they cannot provide per-session token consumption reports, that is itself a red flag.

2. Model the overage. Use the table from Section 1 as a starting framework, adjusted for your group's visit mix.

3. Model the revenue loss. Calculate your G2211 denial rate and CCM denial rate. Multiply by volume. Add the downcoding exposure from truncated MDM narratives.

4. Sum: subscription + overages + revenue loss + implementation = true TCO.

Phase 3: Architecture Evaluation (Week 3)

1. Ask every vendor: "How do you write MDM to the EHR?" If the answer is a single Composition.section.text block, ask the follow-up: "What happens when a clinician edits one problem's assessment after the note is generated?"

2. Ask: "How do you handle G2211 documentation?" If the answer is "the clinician adds it manually," you are paying for an AI scribe that does not scribe the highest-value documentation element in your revenue cycle.

3. Ask: "How do you enforce ICD-10 specificity?" If the answer is "we suggest codes after the note is written," you have a coding tool, not a documentation tool. Specificity must be enforced at capture time.

Phase 4: Proof of Concept (Week 4)

Book a live demo with Scribing.io to run your de-identified notes through our 2026 Token-Cost Simulator with G2211/99487 auto-justification and FHIR R4 write-back. See predictable spend vs. token-cap overages before you switch. The simulator ingests your actual visit complexity distribution and outputs a side-by-side TCO comparison against your current vendor—including the revenue preservation from G2211 and CCM documentation that your current tool is not generating.

The 2026 AI scribe market has matured past the question of "does it save time?" Every ambient tool saves time. The question for a CMIO managing a 20–50 provider internal medicine group with a complex chronic care population is: does it save time while preserving revenue, controlling costs, and producing audit-defensible documentation at maximum ICD-10 specificity?

Token-capped systems cannot answer yes to all four. Scribing.io can.