Gemini hype meets Gemini downtime

This is the kind of tiny failure screenshot that sticks harder than a launch blog post: infrastructure wins on benchmarks, but users remember the moment the assistant flakes out on its own homework.

• –Google’s April 22, 2026 post says Gemini Embedding 2 is generally available via the Gemini API and Vertex AI, positioning it as production-ready multimodal retrieval infrastructure.
• –The screenshot flips that message on its head by showing the Gemini app unable to complete a dead-simple summarization request on Google’s own page.
• –For developers, it highlights the gap between model capability claims and the reliability of the surfaces wrapped around those models.
• –Embeddings are backend plumbing, but the public narrative still gets shaped by front-end UX failures, outages, and rate-limit friction.
• –Gemini does have an established Product Hunt presence, but posts like this show why shipping strong models is only half the job; trust comes from consistent behavior in ordinary use.