{"rewrite":{"id":"r_a6a1bff930d6ac7277ddd6a9","clusterId":"c_3fa743ac385f6eb7ba51cd94","slug":"liquid-ai-releases-pipette-benchmark-app-for-smartphone-ai","model":"deepseek-v4-flash","headline":"Liquid AI Releases Pipette Benchmark App for Smartphone AI","summary":"Liquid AI released Pipette, a free benchmark app for measuring AI execution performance on smartphones. Available on iOS and Android, the iOS version uses the GPU, while Android is CPU-only. Testing on an iPhone 17 Pro with Gemma 4 E2B IT took 18 minutes and results can be exported as CSV.","whyItMatters":"Pipette gives smartphone users a standardized way to measure on-device AI performance across multiple models, with results optionally shared as public data.","webCardHtml":"\u003cp\u003eLiquid AI\u0026#39;s new benchmark app Pipette runs AI models locally on a phone and measures processing speed. The iOS version supports GPU computation, while Android is limited to CPU. A test using Google\u0026#39;s Gemma 4 E2B IT on an iPhone 17 Pro took 18 minutes.\u003c/p\u003e\u003cp\u003eThe app requires account registration and lets users download quantized models from a searchable list. Four test types run by default: end-to-end latency, prefill throughput, decode throughput, and maximum memory usage. Results can be exported as CSV for review in spreadsheet software.\u003c/p\u003e","blueskyPost":"Pipette's iOS version runs on the GPU while Android is CPU-only, so results aren't comparable across platforms. The iPhone 17 Pro run took 18 minutes.","twitterPost":"Pipette's iOS build uses the GPU; Android is CPU-only. Cross-platform scores won't match.","threadsPost":"Pipette splits smartphone AI benchmarking by OS: iOS runs on the GPU, Android on the CPU. That makes scores meaningful only within a platform, not across them. The iPhone 17 Pro test with Gemma 4 E2B IT took 18 minutes, and the app exports CSV for anyone who wants to compare runs.","newsletterBlurb":"Liquid AI has released Pipette, a free benchmark app for measuring AI execution speed on smartphones. Available on iOS and Android, it supports multiple AI models and exports results as CSV. Testing on an iPhone 17 Pro with Gemma 4 E2B IT took 18 minutes.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260825-pipette-phone-ai/\",\"title\":\"I tried using the benchmark app \\\"Pipette\\\" that can measure AI execution performance on your smartphone\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4689,"outputTokens":502,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1788006107,"createdAt":"2026-08-29T12:01:51.000Z","publishedAt":"2026-08-29T12:06:44.000Z","updatedAt":"2026-08-29T12:01:51.000Z"},"cluster":{"id":"c_3fa743ac385f6eb7ba51cd94","canonicalTitle":"自分のスマホのAI実行性能を測定できるベンチマークアプリ「Pipette」を使ってみた","representativeArticleId":"a_b0a8dc8e91bdf447ca8374b8","sourceCount":1,"writtenSourceCount":1,"writeAttempts":1,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[\"Pipette\"],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-08-25T08:22:00.000Z","lastSeenAt":"2026-08-25T08:22:00.000Z","updatedAt":"2026-08-29T12:06:46.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260825-pipette-phone-ai/","title":"自分のスマホのAI実行性能を測定できるベンチマークアプリ「Pipette」を使ってみた"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":["Pipette"],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":null}