v1 API launched

Public release of the humanization API with developer dashboard, API key management, usage tracking, and word-based billing.

100+ languages

Added multilingual support with automatic language detection covering 100+ languages.

Ghost-1 engine

Launched Ghost-1, our first proprietary humanization engine. Significantly improved quality and stealth across all major detectors.

Improved stealth

Tuned output patterns to reduce detection rates on recently updated detectors. Focused on sentence-level variance.

Ghost mode

Added optional input restructuring before humanization, producing a different output structure that increases stealth.

Quality milestone

Achieved 90%+ output quality score through intensive iterations. Stealth evasion rates improved across all major detectors.

Iteration framework

Built a dedicated framework for rapid iteration on output quality and stealth performance. Began tracking quality and stealth metrics per version.

Improved coherence

Better handling of paragraph transitions and logical flow. Reduced cases where humanized output lost the original argument structure.

Markdown preservation

Humanized output now preserves markdown formatting including headers, lists, bold, and paragraph structure from the original input.

Stealth improvements

Reduced repetitive phrasing patterns that were being flagged by newer detector versions. Improved lexical diversity in output.

Long-form text support

Improved chunking and reassembly for texts over 300 words. Better consistency across chunk boundaries.

V2 production release

Major V2 release. New word count handling, improved processing flow, and versioned releases.

V2 engine

Complete rebuild of the core processing system. Designed for higher throughput and better output consistency.

Improved stealth

Addressed detection patterns identified in GPTZero and Originality updates. Improved sentence-level naturalness.

Performance improvements

Reduced average response time and improved throughput for concurrent requests.

Output quality improvements

Reduced hallucinations and meaning drift in humanized output. Better preservation of technical terminology.

Improved coherence

Better handling of academic and technical writing styles. Reduced awkward phrasing in formal contexts.

Stealth update

Retrained on updated detector-aware data. Improved evasion rates across all major detectors.

Expanded vocabulary

Broadened the output vocabulary to reduce word-level repetition and improve naturalness across diverse topics.

Tone and style control

Introduced tone transformation: casual, formal, balanced, and creative. Gives users control over the humanized output style.

V1.1 update

Second iteration with improved coherence for longer texts and reduced repetition artifacts.

Stealth improvements

First round of stealth tuning based on early user feedback. Reduced detection rates on initial detector benchmarks.

Initial release

UndetectedGPT launched with AI text humanization, supporting English text.