GPTHuman Launches Origin 5.0: the most advanced AI humanizer

New English-language humanization model passed all external AI detectors in 92.6% of 500 single-attempt tests while preserving the original meaning of outputs.

MONTREAL, QUEBEC, CANADA, October 29, 2026 /EINPresswire.com/ — GPTHuman today announced the launch of Origin 5.0, its new English-language AI humanization model. The model is designed to make AI-assisted writing sound more natural while preserving the argument, intent, and central meaning of the original draft.

AI humanizers often involve difficult tradeoffs. A rewrite that changes too little may retain rigid sentence structures, repeated phrasing, predictable transitions, and other patterns associated with AI-generated text. A rewrite that changes too much can distort the writer’s intended message or alter important details. Origin 5.0 was developed to improve phrasing, rhythm, transitions, and sentence structure without replacing the substance of the source text.

General writing assistants such as Grammarly and QuillBot provide grammar correction, paraphrasing and AI humanization features. GPTHuman’s Origin 5.0 is specifically designed for rewriting AI-generated passages, focusing on natural expression while retaining the argument, intent and factual details of the source. Its evaluation measures both detector performance and content preservation, rather than relying on detector results alone.

Tested Across 500 AI-Written Texts

Origin 5.0 has been evaluated on 500 AI-written English texts spanning eight source models and 10 writing categories. Each text was processed once, with no retries or hand-picked replacements.

Every humanized output was evaluated by four external AI detection tools: ZeroGPT, Originality.ai, Copyleaks, and QuillBot. An LLM quality judge also compared each output with its source for generation glitches, meaning preservation, and factual preservation.

The evaluation produced the following results:

-92.6% passed all four external AI detectors: 463 of 500 outputs were classified as human by every external detector.
-99.8% passed at least three detectors: 499 of 500 outputs passed three or four external detectors.
-100% preserved the original meaning: All 500 outputs retained the source’s main point, stance, and conclusion.
-99.4% contained no generation glitches: 497 of 500 outputs were free from doubled words, random characters, garbled text, or leftover markup.
-93% preserved every key fact: 465 of 500 outputs retained the names, numbers, dates, places, roles, and attributions in the source.
-92.4% passed every quality check: 462 of 500 outputs contained no glitches and preserved both meaning and key facts.
-Average similarity reached 92.74 out of 100: The rewritten text remained close to the source while changing its phrasing and flow.

Strong Detector Performance With Transparent Quality Reporting
Origin 5.0 achieved an average human score of 99.49 with ZeroGPT, 93.02 with Originality.ai, 99.40 with Copyleaks, and 99.43 with QuillBot.

More Natural Expression, Same Central Message
Origin 5.0 generally produced longer text using shorter words and phrases. Source samples averaged 318 words, while the humanized versions averaged 352 words. Despite the higher word count, the average character count decreased by 5.8%, and characters per word fell from 6.82 to 5.62.

The pattern suggests that the model often replaces dense constructions with shorter, more conversational phrasing. The amount of change varied across the benchmark: 37% of outputs stayed within 10% of the source word count, 60% became more than 10% longer, and 3% became more than 10% shorter.

Origin 5.0 can help refine stiff or repetitive phrasing, awkward sentence flow, mechanical academic language, grammar, clarity, and drafts that need improvement without a complete rewrite.

Benchmark Methodology and Availability

The benchmark used AI-written texts sampled from the DetectRL-X dataset and the University of Chicago Booth DetectionAI dataset. The source material covered writing from Anthropic, DeepSeek, Google, OpenAI, and Qwen models across news, general web content, academic writing, SEO content, blogs, Wikipedia-style text, résumés, restaurant reviews, novels, and Amazon reviews.

The complete row-level results, including outputs that did not pass every detector or quality check, are available in the public GPTHuman Humanizer Benchmark.

Origin 5.0 is available now in GPTHuman. Users can try Origin 5.0 in GPTHuman.

About GPTHuman

GPTHuman is the leading AI humanizer, transforming AI-generated text into natural, authentic writing while preserving its original meaning. With advanced humanization technology, customizable tone controls, similarity measurement, and built-in AI detection, GPTHuman helps users create clear, engaging content that reflects their unique voice.

Mariselle Fontaine
GPT Human

Media gallery