Pangram 3.0 API Migration Guide

Pangram 3.0 API Migration Guide

Pangram's 3.0 API is live! This guide covers everything you need to know about migrating from our v2 endpoints to the new v3 API.

What's New in v3?

We've introduced a third classification category for AI-assisted text. Text that falls into this category is typically written with significant AI and human input—reflecting how people actually use AI tools today.

Key Schema Changes

The v3 API classifies every document with three fraction fields indicating how much of the document falls into each category:

• fraction_human — portion classified as human-written
• fraction_ai_assisted — portion classified as AI-assisted
• fraction_ai — portion classified as AI-generated

These three fields are mutually exclusive and collectively exhaustive; their sum will always equal 1.

Additionally, every document includes three prediction strings at different verbosity levels:

• headline — e.g., "Fully Human Written", "AI Assisted", "AI Detected"
• prediction — detailed explanation of the classification. For example, "We are confident that this document is fully AI-generated" or "We believe that this document is AI-assisted, but not fully AI-generated".
• prediction_short — e.g., "Human", "Mixed", "AI"

Window-Level Analysis

Each input document is broken up into windows. Every window is labeled with:

• label — classification string ('Human', 'Moderately AI-assisted', 'AI Generated', etc.)
• confidence — confidence output from Pangram's detection model ('High', 'Medium', 'Low')
• ai_assistance_score — scalar from 0 to 1 indicating magnitude of AI assistance

Note: Expect additional labels ('Lightly AI-assisted', 'Heavily AI-assisted') to be introduced in the near future.

---

Should I Migrate?

YES if you:

• Use Pangram results in a production environment
• Want improved accuracy on outputs from the latest AI models
• Plan on using the Pangram API beyond April 1, 2026

NO if you:

• Are a researcher who needs a static API for reproducibility

---

Timeline

Date	Event
December 11, 2025	Pangram 3.0 released through the dashboard
January 5, 2026	Pangram 3.0 available via the API
Late January 2026	Pangram 3.1 released (dashboard + API). All 3.0 API users automatically upgraded.
April 1, 2026	Pangram 2.0 APIs sunset

---

V2 Endpoints (Available Until April 1, 2026)

The following v2 endpoints remain available during the transition period:

• text.api.pangram.com
• text-batch.api.pangram.com
• text-sliding.api.pangram.com
• text-extended.api.pangram.com
• dashboard-text.api.pangram.com

---

Field Migration Guide

From text.api.pangram.com

Endpoint: https://text.api.pangram.com → https://text.api.pangram.com/v3

For the most comprehensive understanding of Pangram's predictions, use the top-level fields:

v2 Field	v3 Field	Notes
prediction	headline	Values like "Fully Human Written", "AI Assisted", "AI Detected"
ai_likelihood	fraction_ai	Quantity of text believed to be AI-generated
—	fraction_ai_assisted	New: Quantity believed to be AI-assisted
—	fraction_human	Quantity believed to be human-written

Matching the original endpoint behavior: The original endpoint only used the first window of the text to make its prediction. To match this behavior:

v2 Field	v3 Field
prediction	windows[0].label and windows[0].confidence
ai_likelihood	windows[0].ai_assistance_score

Important: The interpretation of ai_assistance_score differs from ai_likelihood. These fields are not one-to-one. Expect more predictions in the middle range (0–1). Use label to understand our recommended interpretation.

---

From text-batch.api.pangram.com

There is currently no batch endpoint for the v3 API. Use a loop to process each item individually, following the text.api.pangram.com migration guide above.

---

From text-sliding.api.pangram.com

v2 Field	v3 Field	Notes
prediction	headline or prediction	Different verbosity levels available
prediction_short	prediction_short	Unchanged
ai_likelihood, avg_ai_likelihood	fraction_ai, fraction_ai_assisted, fraction_human	Use quantity fields instead of aggregated confidence scores
max_ai_likelihood	fraction_ai (recommended)	Quantity fields preferred

Window changes:

• In text-sliding, windows are overlapping
• In v3, windows are non-overlapping

v2 Window Field	v3 Window Field
windows[].prediction	windows[].label + windows[].confidence
windows[].ai_likelihood	windows[].ai_assistance_score

---

From text-extended.api.pangram.com

Most fields remain unchanged:

Field	Status
headline	Unchanged
prediction	Unchanged
prediction_short	Unchanged
fraction_human	Unchanged
fraction_ai	Unchanged
windows[].label	Unchanged
windows[].confidence	Unchanged
windows[].start_index	Unchanged
windows[].end_index	Unchanged

Fields that changed:

v2 Field	v3 Field	Notes
fraction_mixed	fraction_ai_assisted	Field renamed for clarity
dashboard	public_dashboard_link	Field renamed for clarity
avg_ai_likelihood	fraction_ai, fraction_ai_assisted, fraction_human	Use quantity fields instead of aggregated confidence scores
max_ai_likelihood	fraction_ai, fraction_ai_assisted, fraction_human	Use quantity fields instead of aggregated confidence scores
windows[].ai_likelihood	windows[].ai_assistance_score	AI likelihood and AI assistance scores are not one-to-one. Use label for guidance.

---

Documentation

Full API documentation is available here.

---

Questions?

If you have questions about migrating to v3, please reach out at support@pangram.com.