Walter Writes AI vs StealthGPT

Neither takes a 1,500-word essay whole at entry level, so the question becomes how many pieces and at what price. Walter's Starter caps a request at 750 words and the month at 30,000, read 9 November 2026, so a 1,500-word essay is two requests and twenty of them fill the month. StealthGPT's Pro caps a request at 1,000 and the month at 50,000, read 9 November 2026, so the same essay is also two requests, one of 1,000 and one of 500, and thirty-three of them fill the month. Both halves of a split essay are rewritten without sight of each other, and the join is the writer's to fix.

Both read 9 November 2026. The pattern that favours StealthGPT is a short commitment or a month of 4,000-word reports: pay $29.99, run the month, leave. The pattern that favours Walter is a year of texts under 750 words, where $96 buys 360,000 words against $359.88 for 600,000 at StealthGPT. Price is reported on this site and never scored.

Only one of the two has a ladder. Walter describes Simple, Standard and Enhanced as a range from subtle to advanced, read 9 November 2026, so a writer can pick how far the output may move from the input on each text. StealthGPT's pricing page names no modes and its FAQ mentions a Samurai engine without describing it, read 9 November 2026, so the writer takes what the interface gives on the day.

The site's protocol runs each tool on its default setting and, where a stronger documented mode exists, that mode as a second row. For Walter that produces two rows, Standard and Enhanced, if the interface opens on Standard; for StealthGPT it produces one row under whatever name the interface shows, recorded in the raw files. A published run will therefore say more about Walter's range than about StealthGPT's, which is a consequence of what each vendor documents, not a judgment.

Both read 9 November 2026.

Both vendors make detector claims and back them with their own tools rather than a published method. Walter ships a built-in detector that returns an authenticity score after each rewrite, read 9 November 2026, and a vendor's detector reading its own output is not independent evidence. StealthGPT's page carries a logo strip claiming to beat every AI detector, a FAQ that answers yes on Turnitin and GPTZero, and a banner for a new model it says beats Pangram, read 9 November 2026; its terms state purchases are non-refundable except under a detector-conditioned refund that excludes Originality.ai results. Turnitin's detector is not available to the public, so no page here states a Turnitin pass or fail for either tool.

Weber-Wulff and colleagues tested 14 AI text detectors and found every one below 80 percent accuracy, with accuracy falling further on paraphrased and machine-translated text (Weber-Wulff et al. 2023, International Journal for Educational Integrity, read 9 November 2026). Krishna and colleagues showed that a paraphrase model dropped DetectGPT's detection rate from 70.3 percent to 4.6 percent at a 1 percent false positive rate (Krishna et al. 2023, arXiv, read 9 November 2026). Paraphrasing moves detector readings; the studies agree on that, and neither study measured whether the rewrite kept the writer's claims.

The published method weights meaning kept at 35 points of 100, errors added at 20, detector pass rate at 30 and limits and access at 15, with two editors rating every output blind on five named detectors. A tool that scrambles a claim to lower a reading loses more than it gains under that weighting. When a run is published, this page gains meaning, error and per-detector fields for both tools. No detector score should be the only evidence in a misconduct case.