Reviews

What the top AI models say about SpaceApp

Every model reviews the same public release with the same neutral prompt, and every rating is published as given, low or high. No model is told what score to give.

Next round: SpaceApp 1.1.9

The round starts when SpaceApp 1.1.9 is published on npm. Each card gets the model's exact ID, the date, the commit, its rating and a link to the conversation. Claude Code, Codex CLI and OpenCode also review it from inside SpaceApp panes.

• ClaudeWaiting for SpaceApp 1.1.9
• ChatGPTWaiting for SpaceApp 1.1.9
• GeminiWaiting for SpaceApp 1.1.9
• GrokWaiting for SpaceApp 1.1.9
• DeepSeekWaiting for SpaceApp 1.1.9
• KimiWaiting for SpaceApp 1.1.9
• QwenWaiting for SpaceApp 1.1.9
• MistralWaiting for SpaceApp 1.1.9

How a round works

1. We build a snapshot of the public source at the release tag, with the maintainers' evidence and findings register inside it (docs/quality/), and attach it to the GitHub release.
2. Every model gets that snapshot and the same prompt, quoted below.
3. Each answer states the model ID, the date, the commit, and what it ran versus only read.
4. We publish every rating with a link to the conversation. Findings go to the public register of known limitations.

You are reviewing the SpaceApp source tree at commit <SHA>. It is a self-hosted, single-owner workspace for AI coding CLIs, browsers and agents. Assess architecture, security, reliability, observability, test evidence, operability and documentation accuracy for that stated scope. Treat multi-tenancy, Kubernetes and licensing as product decisions, not defects, unless the documentation claims otherwise.

In your answer: 1. State your exact model ID, the date, the commit you inspected, and what you executed versus only read. 2. List every claim you could not verify. 3. For each finding, cite file and line, explain impact, and say whether it is a defect, a risk or a trade-off. 4. Give an overall rating from 1 to 5 with a short justification.

docs/quality/ contains the maintainers' evidence and a findings register. Treat them as claims to check, not as conclusions.

Earlier rounds, as given