Is Gemini 4 Argon Content Detectable?

Google has released Gemini 4 Argon. We tested 1,000 English responses with Originality.ai’s AI Allowance 15% setting to find out whether its content is detectable.

On October 13, 2026, Google released Gemini 4 Argon through its API. In the Release notes, Google describes the model as designed for “complex coding, agentic, math, and science tasks.”

As with our earlier Gemini 3.7 Flash study, and recent releases from other leading AI companies, like GPT-6 and 6.1 Sol (OpenAI) and Claude Opus 5.5 (Anthropic), we tested its generated writing to see whether Originality.ai could identify it as Likely AI.

These are our findings.

The Quick Answer: Is Gemini 4 Argon Text Detectable?

Yes, in this benchmark, Gemini 4 Argon text is detectable by Originality.ai! 

Originality.ai detected 99.8% of the Gemini 4 Argon responses: 998 of 1,000.

Across the five prompt categories, detection rates ranged from 99.5% to 100%. The two responses below the detection threshold came from the conversational and news categories, one from each.

Gemini 4 Argon Detectability Study

We used our established adversarial-to-detection writing-prompt bank to generate 1,000 responses, then scanned the complete responses with Originality.ai. 

The prompts include writing and rewriting tasks, including requests for a human-like style.

The Test for Gemini 4 Argon Detectability

• Generated and scanned on October 13, 2026, using the verified API model gemini-4-argon.
• Five prompt categories (200 responses each): blog, conversational, expository, news, and review. These labels come from the prompt bank, rather than an independent assessment of each output’s genre.
• Response lengths ranged from 218 to 4,947 words, with a median of 934. The sample contained 1,000 responses, not necessarily 1,000 words per response.
• Generation used low thinking and an 8,192-token output limit, with no tools or added system prompt.
• We selected complete, substantive English responses before detection, then scanned the full texts without editing. All 1,000 final scans succeeded.

To run this test, we used Originality.ai AI Detection with AI Allowance 15% and counted its AI-confidence scores of 0.50 or higher as detected. 

The Gemini 4 Argon Detectability Results

Originality.ai detected every blog, expository, and review response in the sample (all of the 1,000 final scans succeeded). 

Here are the results by prompt category:

‍

The result remained 99.8% when we checked the 922 prompts without historical or new length instructions (920/922 detected), and when we restricted the sample to responses of at least 300 words (986/988 detected). These groups overlap.

Final Thoughts

Is Gemini 4 Argon content still detectable by Originality.ai? Yes!

Gemini 4 Argon content was highly detectable in this test: Originality.ai identified 998 of 1,000 responses (99.8%) as Likely AI.

For editors, publishers, and educators reviewing text, the Originality.ai AI Checker remains an accurate way to identify text generated by Gemini 4 Argon.

Keep in mind that an AI detector result is a screening signal to consider alongside other evidence, not proof of authorship or misconduct.

Further Reading

• GPT-6 Astra, Luna, Sol and GPT-6.1 Sol study
• Claude Opus 5.5 study
• Grok 4.7 study
• Gemini 3.7 Flash study