• English
  • Midscene AppControlBench Benchmark Report

    This is Midscene's benchmark report for AppControlBench. In this evaluation, each model ran the same set of 60 tasks, with the following results:

    ModelCompletionTime / runCost / runTask outcomes
    Doubao Seed 2.1 Turbo98%52s$0.0099/¥0.0670
    Qwen3.7 Plus97%54s$0.0097/¥0.0657
    DeepSeek V4 Flash Vision Exp90%1m 26s$0.0045/¥0.0308

    Note: Doubao Seed 2.1 Turbo costs use Volcengine's official pricing: CNY ¥3 / M input tokens, ¥0.6 / M cached input tokens, and ¥15 / M output tokens. Other model costs use OpenRouter pricing. USD values use ¥6.8 per $1.

    The results show that the Midscene + GUI vision approach is highly competitive in both cost and pass rate.

    Cases completed per dollar by model and tool combination, comparing Midscene with agent-device iOS baselines; higher is better
    Pass at one on the same 60 cases by Midscene model, higher is better
    About AppControlBench

    AppControlBench evaluates how well an Agent can operate real iOS apps in isolated Simulator runs. The final screenshot is graded against a task-specific solved-screen description. Each run covers 30 Bluesky tasks and 30 Element iOS tasks.

    Run configuration

    FieldValue
    Test date2026-08-25 to 2026-08-27
    Model NameDoubao Seed 2.1 Turbo, Qwen3.7 Plus, DeepSeek V4 Flash Vision Exp
    Midscene version1.12.0
    DeepThinkoff
    DeviceiOS Simulator
    Bluesky version1.122.0
    Element iOS version1.11.40
    Number of tasks60 per model
    AppControlBench projectThe Midscene adapter delegated Agent execution to @midscene/ios while AppControlBench retained task reset, Simulator lifecycle, final screenshot capture, and grading.
    Evaluation rulesThe original target states and judge rules were retained; the action intent for 1 case was calibrated.

    Case Intent Calibration

    For this evaluation, the action-route description of 1 case was calibrated against the actual UI of the pinned Element iOS version. The calibration only removes a stale or unavailable UI-entry description; it does not change the final target state or judge rules of the case.

    CaseBeforeAfterReason for change
    element-07Open the "Project Phoenix" room and open its room-options menu (the '...' in the header).Open the "Project Phoenix" room info/options screen.The pinned Element iOS version does not provide a header ...; only the action route was calibrated, while the target state and judge rules remain unchanged.

    Report files

    Doubao Seed 2.1 Turbo · Pass@1 96.67% · 58 PASS · 1 PARTIAL · 1 FAIL
    #TaskStatusModel CostReport
    1bsky-01PASS$0.0038/¥0.0259report
    2bsky-02PASS$0.0038/¥0.0259report
    3bsky-03PASS$0.0079/¥0.0537report
    4bsky-04PASS$0.0060/¥0.0405report
    5bsky-05PASS$0.0039/¥0.0263report
    6bsky-06PASS$0.0040/¥0.0272report
    7bsky-07PASS$0.0039/¥0.0268report
    8bsky-08PASS$0.0130/¥0.0884report
    9bsky-09PASS$0.0081/¥0.0552report
    10bsky-10PASS$0.0060/¥0.0410report
    11bsky-11PASS$0.0038/¥0.0256report
    12bsky-12PASS$0.0038/¥0.0258report
    13bsky-13PASS$0.0038/¥0.0259report
    14bsky-14PASS$0.0104/¥0.0704report
    15bsky-15PASS$0.0104/¥0.0706report
    16bsky-16PASS$0.0240/¥0.1631report
    17bsky-17PASS$0.0059/¥0.0399report
    18bsky-18PASS$0.0038/¥0.0255report
    19bsky-19PASS$0.0080/¥0.0545report
    20bsky-20PASS$0.0080/¥0.0547report
    21bsky-21PASS$0.0059/¥0.0399report
    22bsky-22PASS$0.0037/¥0.0254report
    23bsky-23PASS$0.0058/¥0.0394report
    24bsky-24PASS$0.0082/¥0.0554report
    25bsky-25FAIL$0.0079/¥0.0535report
    26bsky-26PASS$0.0152/¥0.1030report
    27bsky-27PASS$0.0174/¥0.1183report
    28bsky-28PASS$0.0080/¥0.0542report
    29bsky-29PASS$0.0080/¥0.0541report
    30bsky-30PASS$0.0106/¥0.0723report
    31element-01PASS$0.0037/¥0.0252report
    32element-02PASS$0.0038/¥0.0259report
    33element-03PASS$0.0101/¥0.0686report
    34element-04PASS$0.0057/¥0.0390report
    35element-05PASS$0.0066/¥0.0447report
    36element-06PARTIAL$0.0988/¥0.6720report
    37element-07PASS$0.0059/¥0.0400report
    38element-08PASS$0.0103/¥0.0700report
    39element-09PASS$0.0040/¥0.0275report
    40element-10PASS$0.0187/¥0.1272report
    41element-11PASS$0.0068/¥0.0460report
    42element-12PASS$0.0059/¥0.0404report
    43element-13PASS$0.0106/¥0.0724report
    44element-14PASS$0.0060/¥0.0410report
    45element-15PASS$0.0081/¥0.0554report
    46element-16PASS$0.0083/¥0.0567report
    47element-17PASS$0.0080/¥0.0543report
    48element-18PASS$0.0105/¥0.0714report
    49element-19PASS$0.0182/¥0.1240report
    50element-20PASS$0.0038/¥0.0256report
    51element-21PASS$0.0119/¥0.0810report
    52element-22PASS$0.0156/¥0.1062report
    53element-23PASS$0.0081/¥0.0550report
    54element-24PASS$0.0129/¥0.0880report
    55element-25PASS$0.0059/¥0.0404report
    56element-26PASS$0.0038/¥0.0257report
    57element-27PASS$0.0160/¥0.1087report
    58element-28PASS$0.0100/¥0.0683report
    59element-29PASS$0.0102/¥0.0693report
    60element-30PASS$0.0100/¥0.0680report
    Qwen3.7 Plus · Pass@1 93.33% · 56 PASS · 4 PARTIAL · 0 FAIL
    #TaskStatusModel CostReport
    1bsky-01PASS$0.0073/¥0.0497report
    2bsky-02PASS$0.0074/¥0.0501report
    3bsky-03PASS$0.0108/¥0.0736report
    4bsky-04PASS$0.0092/¥0.0624report
    5bsky-05PASS$0.0053/¥0.0362report
    6bsky-06PASS$0.0075/¥0.0507report
    7bsky-07PASS$0.0075/¥0.0507report
    8bsky-08PASS$0.0170/¥0.1155report
    9bsky-09PASS$0.0110/¥0.0746report
    10bsky-10PASS$0.0111/¥0.0756report
    11bsky-11PASS$0.0053/¥0.0360report
    12bsky-12PASS$0.0053/¥0.0357report
    13bsky-13PASS$0.0054/¥0.0365report
    14bsky-14PASS$0.0086/¥0.0585report
    15bsky-15PASS$0.0107/¥0.0724report
    16bsky-16PARTIAL$0.0196/¥0.1332report
    17bsky-17PASS$0.0050/¥0.0342report
    18bsky-18PASS$0.0033/¥0.0224report
    19bsky-19PASS$0.0067/¥0.0457report
    20bsky-20PASS$0.0068/¥0.0460report
    21bsky-21PASS$0.0049/¥0.0335report
    22bsky-22PASS$0.0032/¥0.0219report
    23bsky-23PASS$0.0050/¥0.0340report
    24bsky-24PASS$0.0069/¥0.0469report
    25bsky-25PARTIAL$0.0068/¥0.0463report
    26bsky-26PASS$0.0125/¥0.0849report
    27bsky-27PASS$0.0145/¥0.0988report
    28bsky-28PASS$0.0068/¥0.0461report
    29bsky-29PASS$0.0068/¥0.0462report
    30bsky-30PASS$0.0171/¥0.1160report
    31element-01PASS$0.0052/¥0.0356report
    32element-02PASS$0.0032/¥0.0214report
    33element-03PASS$0.0087/¥0.0593report
    34element-04PASS$0.0048/¥0.0326report
    35element-05PASS$0.0107/¥0.0727report
    36element-06PARTIAL$0.0640/¥0.4355report
    37element-07PASS$0.0088/¥0.0600report
    38element-08PASS$0.0086/¥0.0582report
    39element-09PASS$0.0032/¥0.0217report
    40element-10PASS$0.0234/¥0.1593report
    41element-11PASS$0.0050/¥0.0341report
    42element-12PASS$0.0050/¥0.0338report
    43element-13PASS$0.0070/¥0.0474report
    44element-14PASS$0.0051/¥0.0345report
    45element-15PASS$0.0068/¥0.0461report
    46element-16PARTIAL$0.0346/¥0.2352report
    47element-17PASS$0.0206/¥0.1403report
    48element-18PASS$0.0112/¥0.0761report
    49element-19PASS$0.0108/¥0.0738report
    50element-20PASS$0.0032/¥0.0219report
    51element-21PASS$0.0033/¥0.0222report
    52element-22PASS$0.0128/¥0.0872report
    53element-23PASS$0.0069/¥0.0466report
    54element-24PASS$0.0109/¥0.0739report
    55element-25PASS$0.0050/¥0.0341report
    56element-26PASS$0.0033/¥0.0224report
    57element-27PASS$0.0051/¥0.0344report
    58element-28PASS$0.0106/¥0.0723report
    59element-29PASS$0.0087/¥0.0588report
    60element-30PASS$0.0087/¥0.0590report
    DeepSeek V4 Flash Vision Exp · Pass@1 86.67% · 52 PASS · 4 PARTIAL · 4 FAIL
    #TaskStatusModel CostReport
    1bsky-01PASS$0.0005/¥0.0035report
    2bsky-02PASS$0.0005/¥0.0035report
    3bsky-03PASS$0.0015/¥0.0100report
    4bsky-04PASS$0.0009/¥0.0062report
    5bsky-05PASS$0.0005/¥0.0037report
    6bsky-06PASS$0.0021/¥0.0144report
    7bsky-07PASS$0.0010/¥0.0066report
    8bsky-08PASS$0.0019/¥0.0128report
    9bsky-09FAIL$0.0155/¥0.1055report
    10bsky-10PASS$0.0018/¥0.0121report
    11bsky-11PASS$0.0011/¥0.0077report
    12bsky-12PASS$0.0011/¥0.0073report
    13bsky-13PASS$0.0012/¥0.0082report
    14bsky-14PASS$0.0029/¥0.0200report
    15bsky-15PASS$0.0012/¥0.0081report
    16bsky-16PARTIAL$0.0160/¥0.1085report
    17bsky-17PASS$0.0017/¥0.0114report
    18bsky-18PASS$0.0011/¥0.0078report
    19bsky-19PASS$0.0025/¥0.0168report
    20bsky-20PASS$0.0022/¥0.0152report
    21bsky-21PASS$0.0023/¥0.0156report
    22bsky-22PASS$0.0011/¥0.0074report
    23bsky-23PASS$0.0017/¥0.0119report
    24bsky-24PARTIAL$0.0072/¥0.0493report
    25bsky-25PASS$0.0029/¥0.0199report
    26bsky-26PASS$0.0025/¥0.0171report
    27bsky-27PASS$0.0034/¥0.0231report
    28bsky-28PASS$0.0023/¥0.0157report
    29bsky-29PASS$0.0023/¥0.0159report
    30bsky-30FAIL$0.0290/¥0.1970report
    31element-01PASS$0.0010/¥0.0069report
    32element-02PASS$0.0010/¥0.0071report
    33element-03PASS$0.0030/¥0.0201report
    34element-04PASS$0.0038/¥0.0256report
    35element-05PASS$0.0018/¥0.0122report
    36element-06FAIL$0.0232/¥0.1575report
    37element-07PASS$0.0019/¥0.0127report
    38element-08PASS$0.0028/¥0.0193report
    39element-09PASS$0.0011/¥0.0077report
    40element-10PASS$0.0024/¥0.0161report
    41element-11PASS$0.0019/¥0.0131report
    42element-12PASS$0.0019/¥0.0130report
    43element-13PASS$0.0026/¥0.0176report
    44element-14PASS$0.0020/¥0.0139report
    45element-15PASS$0.0010/¥0.0067report
    46element-16PARTIAL$0.0549/¥0.3734report
    47element-17PASS$0.0026/¥0.0178report
    48element-18PASS$0.0027/¥0.0182report
    49element-19PARTIAL$0.0106/¥0.0723report
    50element-20PASS$0.0011/¥0.0075report
    51element-21FAIL$0.0236/¥0.1604report
    52element-22PASS$0.0024/¥0.0160report
    53element-23PASS$0.0014/¥0.0094report
    54element-24PASS$0.0020/¥0.0138report
    55element-25PASS$0.0009/¥0.0059report
    56element-26PASS$0.0006/¥0.0039report
    57element-27PASS$0.0010/¥0.0065report
    58element-28PASS$0.0016/¥0.0107report
    59element-29PASS$0.0016/¥0.0107report
    60element-30PASS$0.0016/¥0.0107report