Skip to content
Documentation menu

Share your model · Android

Test your phone with Ompera Bench.

See how fast models run on your Android phone and how hot it gets.

Download Ompera Bench
  • Android 12+, ARM64
  • Wi-Fi for downloads
  • Fairer unplugged
Qwen3 1.7BExecuTorch · CPU · Big cores
20.2answer, tok/s
77.6reading, tok/s
CPU 61°Case 34°
Run 10 of 18

  1. Install the app

    Open the downloaded APK on your phone. Android asks once to allow installs from your browser and may show a Play Protect notice.

    Ompera Bench opens with your phone's chip, cores and memory at the top.

  2. Choose how much to test

    • Quick 1 run of one model Time~1 min Space1–3.7 GB
    • One model ~20 runs: every compute unit and core group Time~25 min Space1–3.7 GB
    • Everything ~180 runs: every model, everywhere it runs For the ranking Time~3.5 h Space~18 GB ~7 GB on a 6 GB phone, 21 GB+ on 12 GB
    • Custom Pick models, units, cores and tests, including long input TimeYou choose SpaceYou choose

    For an 8-core phone with 8 GB of memory, without the cool-down between runs. The app shows the exact time and download size before you start. Stop whenever you like: Everything continues where it stopped.

    A model on an enginellama.cppLiteRT-LMExecuTorchMNNGenieX
    Compute unitCPUGPUNPU
    CPU coresPrimeBigSmallAllApp default

    Each engine runs only where it supports the phone; GenieX needs a supported Snapdragon NPU. Models come from their publishers on first use. Open-source licenses

  3. Run it and watch

    Pause sharing in the Ompera app, then tap Run. The phone cools down between runs, so they stay comparable.

    • Done
    • Slow or stopped
    • Failed: tap it for the reason
    • Running
    • Waiting
    Stop at any time; finished runs are kept.

  4. Compare the results

    Results shows every run on your phone. Ranking lists every phone, with its fastest result per model and how it got it.

    Results · Qwen3 1.7B · answer speed
    • ExecuTorch · CPU20.2
    • LiteRT-LM · GPU16.8
    • MNN · CPU14.1
    • llama.cpp · CPU11.0
    Example. Also as a table, by input length, or against heat.
    Ranking · every phone, tok/s
    DeviceSoC · RAMQwen3 1.7BGemma 4 E2B
    Phone ASoC A · 12 GB40.2LiteRT-LM28.1LiteRT-LM
    Phone BSoC B · 8 GB20.2ExecuTorch—Not tested
    Your phoneSoC · 8 GB11.0llama.cpp—Not tested
    Example. Under each phone's name, what it has: B fastest cores, M middle, L slowest, G GPU, N NPU. In a cell, the blocks that did the work are filled. A bar is the speed against the fastest. Tap a model to sort.
Sent to Ompera
  • Phone model and chip
  • Settings, speeds and temperatures of each run
  • Error reports when something fails
Never sent
  • Prompts and answers
  • Your account
  • Your location

Switch both off in the app's Settings. Privacy

How do I update it?

Open Settings in the app and tap Check for updates: it downloads the new version, checks its size, checksum and signature, and opens Android's installer. The app also checks by itself and shows Update. Android asks once whether Ompera Bench may install its own updates.

Something failed? The app sends an error report by itself. To tell us more, use Feedback.