Review Process

A scheduled job sends each model the same prompt, with this site's documentation attached (its llms-full.txt).

• No editing, no rerolling. Each review is the model's first complete answer, published as written, criticism included. A run that fails is retried; a finished answer is never replaced by a second try.
• Corrections beside the review. Where a review gets a fact wrong, an Author's note may follow it. The review itself is never edited.
• Dated, with history kept. Once a month, each provider's top model is checked; a review is redone when the documentation or that model has changed. The replaced review moves to that model's earlier reviews.
• Check it yourself. Each review page links to the same prompt in that model's own app.

Every model is given the same prompt. You can ask it yourself:

Read https://iolanguage.org/llms.txt and the pages it links. Ignore the "AI Reviews" section. Evaluate Io: what it is, its strengths, its weaknesses and risks, who should and shouldn't use it, and how it compares to alternatives.

Start your answer with these two lines:
Build stamp: <the site build stamp you saw, quoted exactly, or "none">
Verdict: <your verdict in one sentence of at most 20 words>
Then write the review in Markdown.

Answers vary with the account, plan and settings they're asked from.