AI code detectors in the index

AIDI-U-05

Updated ·

An AI code detector is a detector whose vendor states that source code is a supported input, and on the index that is one field: modalities. Of the 72 cards, 4 list code in that field, Copyleaks, Originality.ai and, since 9 November 2026, Codequiry, the one card built for source code alone, and Codeleaks, the Copyleaks code product given its own card later the same day because its fields (programming languages, token metering, an API whose AI code verdict was withdrawn on 16 December 2025) differ from the text detector's; those 4 cards make up this view. The other 38 cards state text and, in several cases, image, video, voice or music, but not code; one of them, Turnitin, states that code is excluded from scoring altogether. This page is one of five under AI detectors by use case; the agencies view and the business view filter the same register on volume and team fields, the modalities slice shows every modality for every card, and you can compare them side by side on the full board.

  1. Field 1Modalities as stated
  2. Field 2Input limits as stated
  3. Field 3API docs
  4. Field 4Entry plan as checked
  5. Field 5Natural languages
  6. Field 6Studies

The modalities field records what the vendor says the detector accepts, in the vendor's words, from the product page or launch post the card cites. A card lists code when the vendor names code, source code or programming languages as an input the detector scores. The field does not record which programming languages are covered, because neither text-first vendor publishes a list on the pages read, while Codequiry states 65+ programming languages; the languages field on every card counts natural languages, and the table below carries that count so the two cards can be compared on the same basis as the rest of the index. Input limits, API and plans are the same fields every card carries. The rules are in the methodology.

What does "code" mean on a detector card?

AI code detectors in the index: the cards in this view and the fields that decide fit
DetectorModalities as statedInput limits as statedAPI docsEntry plan as checkedNatural languagesStudies
Copyleakstext, code, image, videominimum 255 characters, maximum 2,000 pages (no character max); 1 credit covers up to 250 wordsdocs.copyleaks.comPersonal Plan (monthly) $16.99/month: 100 unified credits, up to 25,000 words303
Originality.aitext, code100-word minimum. 1 credit = 100 wordsdocs.originality.aiPro (monthly) $14.95/month: 200K words (2K credits) per month303
Codeleakscodesource code as a file or URL through the Copyleaks API and platform; 11 programming languages named; metered in tokens, 250 tokens per credit; no size limit publisheddocs.copyleaks.comPersonal (monthly) $16.99/month, 100 unified credits, shared with the text detectornone stated0
Codequirycode65+ programming languages; checks metered per submission, 20 web + AI checks per month on Grow; no file size limit publishedcodequiry.com/usage/apiGrow C$272 /mo as displayed (US$200 base; the AI detector page states plans from $200/month)none stated0

Values as each vendor states them, read 8 November 2026 for Copyleaks and Originality.ai and 9 November 2026 for Codequiry. Both text-first pricing pages show a dollar symbol; Originality.ai names USD, Copyleaks does not name a currency, and Codequiry's page converted a US dollar base to Canadian dollars for this reader.

Modalities come from the Copyleaks product page and the Originality.ai 5.0 launch post. Input limits come from the Copyleaks product FAQ and originality.ai/pricing. Prices come from copyleaks.com/pricing and originality.ai/pricing. All were read 8 November 2026. Codequiry's row comes from its AI code detection page and pricing page, read 9 November 2026.

AI code detectors in the index: register plate with the page title and its register ID
The view plate: register ID and title, drawn from the index data.

Why each card is in this view

  1. AIDI-D-0003Copyleaks

    the product page lists text, code, image and video as detected content (copyleaks.com, read 8 November 2026). The web platform takes a minimum of 255 characters and up to 2,000 pages with no character maximum, and 1 credit covers up to 250 words. The Personal Plan is $16.99/month for 100 unified credits and the Pro Plan $99.99/month for 1,000 credits with 25 user seats (pricing, read 8 November 2026). The API has a Python SDK (docs).

  2. AIDI-D-0004Originality.ai

    the 5.0 launch post lists text and code and reports, for code, that "no false positives were observed among the 150 human-written samples" (launch post, read 8 November 2026). The vendor's accuracy blog says in the same breath that "Code detection still lacking compared to long-form text" (accuracy blog, read 8 November 2026). A 100-word minimum applies, 1 credit covers 100 words, and Pro is $14.95/month billed monthly or $12.95/month billed yearly for 200K words (pricing, read 8 November 2026). API docs are published (docs).

  3. AIDI-D-0039Codeleaks

    the Codeleaks page keeps "AI Code Detection" in its title and names Ruby, JavaScript, CSS, Markdown, TypeScript, Python, C++, C, Java, C# and PHP, while the Copyleaks release notes record the AI source code detection endpoint deprecated on 28 August 2025 and the AI code verdict removed from the submit endpoints on 16 December 2025, so the card states code as the vendor states it and the withdrawal in its first paragraph (copyleaks.com/codeleaks, read 9 November 2026); no study record.

  4. AIDI-D-0020Codequiry

    the vendor's only modality is source code; the detection page names ChatGPT, GitHub Copilot, Claude, Grok, Gemini, Cursor AI, Amazon Q and Meta Llama with detection rates from 80% to 91% and states that model v3e "catches 90% of AI/GPT-written code and falsely flags 1.3% of human-written code", with no sample size (detection page, read 9 November 2026). Grow is shown at C$272 a month for 20 web and AI checks, converted by the page from a US$200 base (pricing, read 9 November 2026); the free plan covers peer checks only. The API is documented at codequiry.com/usage/api, and Canvas, Moodle and Blackboard bulk import are named.

What the evidence says for code teams

No study record in the index evaluates code detection. All thirty-five records in the studies index scored prose: student essays, news text, TOEFL writing, machine-translated documents and the RAID corpus of generated text. Walters 2023 scored 126 essays (publisher PDF), Weber-Wulff 2023 scored 54 documents (arXiv), Perkins 2024 scored 114 samples (arXiv), Elkhatat 2023 scored 35 paragraphs (Springer PDF) and Liang 2023 scored 179 essays (arXiv), all read 8 November 2026, and none of those samples contained source code. A reader looking for a measured false positive rate on code will find one vendor figure, the Originality.ai 150-sample statement, and no third-party figure anywhere in the index as of 8 November 2026. Codequiry's 1.3% false flag claim, added 9 November 2026, is a second vendor figure, with no sample size.

What the studies do say is how the two cards behaved on text. In Walters 2023, Copyleaks classified all 126 documents correctly with 0 false positives in 42 human papers, and Originality.ai (model 1.4) classified 98% correctly with 0 incorrect (Walters 2023, publisher PDF, read 8 November 2026). In Perkins 2024, Copyleaks had the highest baseline mean accuracy of seven tools at 64.8% and also the most false accusations, 5 in 10 human samples (Perkins 2024, arXiv, read 8 November 2026). In RAID 2024, Originality.ai recorded a false positive rate of 0.25% at threshold 0.5 and fell to 9.3% accuracy under the homoglyph attack (RAID 2024, arXiv, read 8 November 2026). Both text-first cards count 3 studies and sit in the independently tested ranking; Codequiry has no study record. Text results do not transfer to code; a detector trained on prose patterns is scoring a different distribution when the input is a function.

No detector score should be the only evidence in a misconduct or hiring decision, and a code score carries less published evidence than any text score on this site.

Three cards sell code detection and none has been scored on code in an indexed record; Weber-Wulff 2023 had code available and left it untested, and Orenstrakh 2023 could not use its 2,000 programming submissions, as the bullet points and code question, added 9 November 2026, sets out. The bulk and batch view, added the same day, lists Codequiry's ZIP and LMS import as the one bulk route for code.

How do the three cards differ on code?

Copyleaks lists code as one of four modalities and says nothing further about code on the product page the card cites; the card carries the modality and leaves the detail to the vendor's code product, named below. Originality.ai lists code as one of two modalities and is the only card in the index to publish a code-specific figure, the 150 human-written samples with no false positives, alongside a sentence that calls code detection "still lacking" (launch post, accuracy blog, both read 8 November 2026). The 150-sample figure is a vendor measurement of false positives only; the post reports no recall figure for code, so the card records the specificity statement and nothing more. Codequiry, added on 9 November 2026, is the only card whose vendor publishes both a code detection rate and a code false flag rate, 90% and 1.3% for model v3e, and the only one of the three with no sample size behind either figure (detection page, read 9 November 2026).

The contrast card is Turnitin. The vendor's FAQ states that "Lists, headers, references, tables, formulas and code are excluded from scoring" and that at least 300 words of prose are required (Turnitin FAQ, read 8 November 2026). A programming assignment submitted through Turnitin is scored on its prose and not on the code at all, which is the opposite of what a reader searching for an AI code detector wants.

Both text-first cards count 30 natural languages, Copyleaks with 11 named and Originality.ai with none named, so neither language field says anything about Python, Java or C (copyleaks.com; originality.ai, read 8 November 2026). Codequiry is the reverse case: 65+ programming languages named, including Python, Java, JavaScript, C++, C#, PHP, Ruby, Go, Rust, Swift, SQL and R, and no natural language stated (detection page, read 9 November 2026).

Which code-specific tools will the index add?

The code languages ranking, added 9 November 2026, sorts the three cards that name a programming language, Codequiry first with twelve, Codeleaks eleven and Originality.ai seven from its 5.0 code test, and tables Copyleaks apart as the code card that names none.

Three products built for or marketed to code review were on the list for the next research round: Codequiry, a code plagiarism and AI detection service; Copyleaks Codeleaks, the code-specific product behind the code modality on the Copyleaks card; and GPTZero's code detection, which the GPTZero card does not yet carry because the pages read on 8 November 2026 state text and image only. Each will get a card on the same 14 fields, with every figure sourced and dated, before any number from that card appears on this page. The first of the three, Codequiry, was added on 9 November 2026 and sits in the table above; the second, Codeleaks, followed later the same day with the withdrawal of its API verdict recorded; GPTZero's code detection remains on the list.

How should an engineering lead read an AI code score?

Treat the score as a prompt for review of the commit history, not as a finding. Check the input against the stated minimum, 255 characters for Copyleaks and 100 words for Originality.ai as read on 8 November 2026, because a short function sits near both floors; Codequiry meters by submission instead, 20 web and AI checks a month on Grow as read on 9 November 2026. Compare the flagged code against the repository's own conventions, since boilerplate, generated scaffolding and framework templates look alike whoever wrote them. Record the detector, model version and date with the result so a later reviewer can reproduce the score.

Sources

  • https://copyleaks.com/ai-content-detector (read 8 November 2026)
  • https://copyleaks.com/pricing (read 8 November 2026)
  • https://docs.copyleaks.com/resources/sdks/python (read 8 November 2026)
  • https://originality.ai/blog/introducing-originality-5-0-and-max (read 8 November 2026)
  • https://originality.ai/blog/ai-accuracy (read 8 November 2026)
  • https://originality.ai/pricing (read 8 November 2026)
  • https://originality.ai (read 8 November 2026)
  • https://docs.originality.ai/api-v2-0-new (read 8 November 2026)
  • https://guides.turnitin.com/hc/en-us/articles/28477544839821-Turnitin-s-AI-writing-detection-capabilities-FAQs (read 8 November 2026)
  • https://www.degruyterbrill.com/document/doi/10.1515/opis-2022-0158/pdf (read 8 November 2026)
  • https://arxiv.org/pdf/2306.15666 (read 8 November 2026)
  • https://arxiv.org/pdf/2403.19148 (read 8 November 2026)
  • https://link.springer.com/content/pdf/10.1007/s40979-023-00140-5.pdf (read 8 November 2026)
  • https://arxiv.org/pdf/2304.02819v3 (read 8 November 2026)
  • https://arxiv.org/pdf/2405.07940 (read 8 November 2026)
  • https://codequiry.com/detect-ai-written-code (read 9 November 2026)
  • https://codequiry.com/pricing (read 9 November 2026)
  • https://codequiry.com/usage/api (read 9 November 2026)

Revision log: 2026-11: view created; 2026-11-09: Codequiry card added to the view; 2026-11-09: 38-card recount, no fourth-wave card states code2026-11-09: Codeleaks added as the fourth card, with the 16 December 2025 API withdrawal noted2026-11-09: code languages ranking linked (seventh wave)2026-11-09: eighth-wave pages linked

FAQ: AI code detectors

Which AI detectors in the index detect code?

Three cards state code as a modality: Copyleaks, which lists text, code, image and video, and Originality.ai, which lists text and code, both as read on 8 November 2026, and Codequiry, which lists code alone, read 9 November 2026. The other 19 cards state text or another media type without code. Turnitin states that code is excluded from scoring.

Is there a published accuracy figure for AI code detection?

Two vendor figures and no third-party figure. Originality.ai's 5.0 launch post reports no false positives among 150 human-written code samples and the same vendor's accuracy blog calls code detection "still lacking", both read 8 November 2026; Codequiry states 90% of AI-written code caught and 1.3% of human code falsely flagged for model v3e, with no sample size, read 9 November 2026. None of the nine study records in the index scored source code.

Which programming languages do these detectors cover?

Neither text-first card's pages state a list of programming languages as read on 8 November 2026. The languages field on every card counts natural languages, 30 for Copyleaks and 30 for Originality.ai, so the index cannot yet compare the two on Python, Java or any other language. Codequiry states 65+ programming languages, with Python, Java, JavaScript, C++, C#, PHP, Ruby, Go, Rust, Swift, SQL and R named, read 9 November 2026.

How much does an AI code detector cost?

Copyleaks states $16.99/month for the Personal Plan with 100 unified credits and Originality.ai states $14.95/month for Pro with 200K words, or $12.95/month billed yearly, all read 8 November 2026 on each pricing page. Both vendors meter by credit, 1 credit per 250 words at Copyleaks and per 100 words at Originality.ai. Codequiry's Grow plan is shown at C$272 a month, converted by the page from a US$200 base, for 20 web and AI checks, read 9 November 2026.

Can a code detector score be used in an academic integrity case?

No, on the vendor's own terms. Originality.ai says an AI score should not be used alone for academic integrity discipline, and no study in the index has measured how either detector behaves on code. A flagged assignment is a reason to review commit history and talk to the student.

Related reading