{
  "text": "7月31號,Deepseek发布了VisFlash正式版,這是它史上最強的模型。一周之後的今天,Deepseek給全部的API用戶都發了一個通知,計劃近期整體上調價格,而且預計漲幅較大。我第一反應是看錯了,因為Deepseek在文印象裡面一直只發降價的新聞。我翻了一下,最接近漲價的一次是今年6月的上旬,風股的定價,高峰時段的雙倍的價格,僅此而已。所以很多小伙伴來問,這次憑什麼漲,是要收割了嗎?所以我來跟大家分析一下。先說它原來有多便宜,真的非常便宜。\n很多人沒用過Deepseek的按量集費,所以沒有這個感覺。我們先來看單價,就是每百萬Token的輸入輸出,Flash的價格是一塊和兩塊,Pro是三塊和六塊,在國產模型裡面就是偏低的。然後它還有兩個別家沒有的東西,第一個就是一個million的100萬的上下文,不額外收費。第二個就是緩存命中只有輸入價格的五十分之一,相當於只有兩分錢,遠低於其他的廠商。而且還有一個細節,就是Deepseek的緩存優化做得非常好,寫代碼的場景裡面,命中率基本上95%以上。這樣算下來,其實\n100萬Token的綜合單價只要六分錢,1億Token只要六塊錢。前幾天7月31號,V4Flash正式版本一發,能力大幅度增強。光看跑分已經超過了,參數量比它大好幾倍的JRM5.2。又便宜,然後又強。關鍵的它還快。\n關鍵的它還快,土字速度大概在每秒100個Token以上,甩開其他大模型一大截。所以跑那些agent任務的時候特別絲滑,便宜,能力又強,速度又快,三樣都湊起來。可以說在某種程度上來說,打破了一個不可能的三角。結果就是API的調用量翻了好幾倍,官方沒公布具體的調用數據,但可以從它最大的客戶OpenCode上去推算。\nOpenCode不光是個agent工具,它還提供這種大模型的調用的服務,裡面的DeepSeek就是官方提供的。套餐會有一些折扣,所以比DeepSeek官方去安量收費要便宜一點。而且它又是個開源的項目,所以它不限制去接什麼其他的工具。所以很多人,包括我在內,都從OpenCode裡面去使用DeepSeek。OpenCode裡面提供免費的V4Flash,付費套餐Go裡面也有收費版。我自己訂了這個套餐,實際用下來,免費版跟收費版其實沒有太大差別,只是免費版它有調用次數的限制。然後接著我們來看具體的數字。OpenCode的工具,\n官方說,8月1日,V4Flash發布的第二天,免費版燒了5T的Token,收費版燒了3T的Token,第二天就是8T。這還只是開始,OpenCode有一個實時的數據頁面,收費Token的日消耗量。8月5日已經漲到了6.3T,是Flash正式版發布前的4倍。如果我們再去按照免費跟收費版的比例來推算一下,免費部分一天還要再燒10.5T左右的Token,加起來一天大概要燒17T左右的Token。17T是個什麼概念?就是OpenRoot,我們知道它是全球最大的模型的主軟,\n基本上沒有質疑,幾百個大模型,800萬的用戶,全平台加起來一天也就3T多的Token。OpenCode在DeepSeek一個模型上就要燒的量,是整個OpenRooter幾百個模型用量的總和的5倍。所以我們來推測一下,官方沒有說漲價的原因,但我認為最有可能的解釋就是,算力真的不太夠了。於是Flash變強之後,吸引來的是Agent的重度用戶,工具一接上之後,Token就是沉吊的燒起來了。光是OpenCode這一個客戶一天就能干到17T,用量一周直接漲了4倍。\n但是背後這個卡這個算力不會一周之內就多出4倍,能立刻的去見效。所以只能拿價格去來做暫時的調節。好了,接下來趕緊跟大家來分享一下快訊。然後,今天視頻就到這裡,我是Dizon黑心李少,我們下次見。\n謝謝。",
  "language": "zh",
  "duration": 195.1405,
  "segments": [
    {
      "id": 0,
      "seek": 0,
      "start": 0.0,
      "end": 29.32,
      "text": "7月31號,Deepseek发布了VisFlash正式版,這是它史上最強的模型。一周之後的今天,Deepseek給全部的API用戶都發了一個通知,計劃近期整體上調價格,而且預計漲幅較大。我第一反應是看錯了,因為Deepseek在文印象裡面一直只發降價的新聞。我翻了一下,最接近漲價的一次是今年6月的上旬,風股的定價,高峰時段的雙倍的價格,僅此而已。所以很多小伙伴來問,這次憑什麼漲,是要收割了嗎?所以我來跟大家分析一下。先說它原來有多便宜,真的非常便宜。",
      "tokens": [
        50365,
        22,
        6939,
        12967,
        18616,
        11,
        11089,
        595,
        405,
        916,
        28926,
        34688,
        2289,
        53,
        271,
        25680,
        1299,
        15789,
        27584,
        42096,
        11,
        16600,
        11284,
        45399,
        5708,
        8661,
        22752,
        1546,
        41908,
        39823,
        1543,
        2257,
        37390,
        26362,
        1546,
        12074,
        11,
        11089,
        595,
        405,
        916,
        17798,
        38714,
        1546,
        4715,
        40,
        9254,
        1486,
        114,
        7182,
        14637,
        2289,
        8990,
        19550,
        6498,
        11,
        30114,
        5087,
        225,
        17463,
        16786,
        27662,
        23987,
        5708,
        32943,
        45441,
        30921,
        11,
        22942,
        30156,
        30114,
        14065,
        110,
        3509,
        227,
        23915,
        3582,
        1543,
        1654,
        18049,
        22138,
        20481,
        1541,
        4200,
        13133,
        2289,
        11,
        11471,
        11089,
        595,
        405,
        916,
        3581,
        17174,
        35825,
        45007,
        32399,
        34448,
        14003,
        14637,
        47421,
        45441,
        1546,
        41002,
        1543,
        1654,
        42716,
        2289,
        8861,
        11,
        8661,
        14468,
        17463,
        14065,
        110,
        45441,
        1546,
        27505,
        1541,
        38937,
        21,
        6939,
        1546,
        5708,
        4479,
        105,
        11,
        22713,
        14356,
        94,
        1546,
        12088,
        45441,
        11,
        12979,
        26555,
        108,
        6611,
        28427,
        1546,
        6306,
        247,
        35477,
        1546,
        45441,
        30921,
        11,
        9502,
        227,
        17947,
        48420,
        1543,
        7239,
        20778,
        7322,
        7384,
        247,
        7384,
        112,
        3763,
        11361,
        11,
        2664,
        9487,
        29713,
        239,
        7598,
        14065,
        110,
        11,
        1541,
        4275,
        18681,
        5935,
        110,
        2289,
        7434,
        30,
        7239,
        1654,
        3763,
        47349,
        6627,
        7360,
        238,
        8861,
        1543,
        10108,
        4622,
        11284,
        19683,
        3763,
        2412,
        6392,
        27364,
        2415,
        250,
        11,
        8034,
        14392,
        27364,
        2415,
        250,
        1543,
        51831
      ],
      "temperature": 0,
      "avg_logprob": -0.133754112870221,
      "compression_ratio": 1.289237668161435,
      "no_speech_prob": 8.627482409040432e-12,
      "source": "whisper"
    },
    {
      "id": 1,
      "seek": 2932,
      "start": 29.32,
      "end": 59.32,
      "text": "很多人沒用過Deepseek的按量集費,所以沒有這個感覺。我們先來看單價,就是每百萬Token的輸入輸出,Flash的價格是一塊和兩塊,Pro是三塊和六塊,在國產模型裡面就是偏低的。然後它還有兩個別家沒有的東西,第一個就是一個million的100萬的上下文,不額外收費。第二個就是緩存命中只有輸入價格的五十分之一,相當於只有兩分錢,遠低於其他的廠商。而且還有一個細節,就是Deepseek的緩存優化做得非常好,寫代碼的場景裡面,命中率基本上95%以上。這樣算下來,其實",
      "tokens": [
        50365,
        20778,
        4035,
        4913,
        9254,
        8816,
        11089,
        595,
        405,
        916,
        1546,
        26613,
        26748,
        26020,
        50071,
        11,
        7239,
        6963,
        6287,
        31742,
        1543,
        5884,
        10108,
        3763,
        4200,
        24227,
        45441,
        11,
        5620,
        23664,
        31906,
        14099,
        51,
        8406,
        1546,
        12290,
        116,
        14028,
        12290,
        116,
        7781,
        11,
        25680,
        1299,
        1546,
        45441,
        30921,
        1541,
        2257,
        26268,
        12565,
        16313,
        26268,
        11,
        12681,
        1541,
        10960,
        26268,
        12565,
        30566,
        26268,
        11,
        3581,
        8053,
        33299,
        41908,
        39823,
        32399,
        5620,
        7437,
        237,
        41377,
        1546,
        1543,
        10213,
        11284,
        15569,
        34623,
        16158,
        5155,
        6963,
        1546,
        26978,
        11,
        40760,
        5620,
        8990,
        40388,
        1546,
        6879,
        14099,
        1546,
        5708,
        4438,
        17174,
        11,
        1960,
        10178,
        235,
        12022,
        18681,
        50071,
        1543,
        49085,
        5620,
        12608,
        102,
        39781,
        25236,
        5975,
        35244,
        12290,
        116,
        14028,
        45441,
        30921,
        1546,
        21001,
        20145,
        6627,
        9574,
        2257,
        11,
        15106,
        13118,
        19488,
        35244,
        16313,
        6627,
        29483,
        11,
        34489,
        41377,
        19488,
        9572,
        31309,
        17527,
        254,
        45581,
        1543,
        22942,
        15569,
        8990,
        41588,
        27694,
        11,
        5620,
        11089,
        595,
        405,
        916,
        1546,
        12608,
        102,
        39781,
        12827,
        103,
        23756,
        10907,
        5916,
        14392,
        2131,
        11,
        4510,
        104,
        19105,
        16337,
        120,
        1546,
        21446,
        50218,
        32399,
        11,
        25236,
        5975,
        44866,
        37946,
        5708,
        15718,
        4,
        29497,
        1543,
        8377,
        19497,
        27769,
        11,
        14139,
        51865
      ],
      "temperature": 0,
      "avg_logprob": -0.08397591807854869,
      "compression_ratio": 1.3431151241534989,
      "no_speech_prob": 1.7579502090137034e-11,
      "source": "whisper"
    },
    {
      "id": 2,
      "seek": 5932,
      "start": 59.32,
      "end": 73.32,
      "text": "100萬Token的綜合單價只要六分錢,1億Token只要六塊錢。前幾天7月31號,V4Flash正式版本一發,能力大幅度增強。光看跑分已經超過了,參數量比它大好幾倍的JRM5.2。又便宜,然後又強。關鍵的它還快。",
      "tokens": [
        50365,
        6879,
        14099,
        51,
        8406,
        1546,
        9261,
        250,
        14245,
        24227,
        45441,
        14003,
        4275,
        30566,
        6627,
        29483,
        11,
        16,
        25459,
        51,
        8406,
        14003,
        4275,
        30566,
        26268,
        29483,
        1543,
        8945,
        23575,
        6135,
        22,
        6939,
        12967,
        18616,
        11,
        53,
        19,
        25680,
        1299,
        15789,
        27584,
        42096,
        8802,
        2257,
        14637,
        11,
        8225,
        13486,
        3582,
        3509,
        227,
        13127,
        24228,
        252,
        22752,
        1543,
        20690,
        4200,
        32585,
        6627,
        25338,
        19869,
        8816,
        2289,
        11,
        47375,
        30622,
        26748,
        11706,
        11284,
        3582,
        2131,
        23575,
        35477,
        1546,
        41,
        49,
        44,
        20,
        13,
        17,
        1543,
        17047,
        27364,
        2415,
        250,
        11,
        10213,
        17047,
        22752,
        1543,
        14899,
        41668,
        113,
        1546,
        11284,
        7824,
        10251,
        1543,
        51065
      ],
      "temperature": 0,
      "avg_logprob": -0.17910497328814337,
      "compression_ratio": 1.0635593220338984,
      "no_speech_prob": 2.2023272094884305e-11,
      "source": "whisper"
    },
    {
      "id": 3,
      "seek": 7332,
      "start": 73.32,
      "end": 93.32,
      "text": "關鍵的它還快,土字速度大概在每秒100個Token以上,甩開其他大模型一大截。所以跑那些agent任務的時候特別絲滑,便宜,能力又強,速度又快,三樣都湊起來。可以說在某種程度上來說,打破了一個不可能的三角。結果就是API的調用量翻了好幾倍,官方沒公布具體的調用數據,但可以從它最大的客戶OpenCode上去推算。",
      "tokens": [
        50365,
        14899,
        41668,
        113,
        1546,
        11284,
        7824,
        10251,
        11,
        45506,
        22381,
        31217,
        13127,
        32044,
        3581,
        23664,
        20956,
        6879,
        3338,
        51,
        8406,
        29497,
        11,
        3526,
        102,
        8949,
        9572,
        5000,
        3582,
        41908,
        39823,
        2257,
        3582,
        1486,
        103,
        1543,
        7239,
        32585,
        4184,
        13824,
        559,
        317,
        26443,
        39517,
        20643,
        38446,
        6948,
        110,
        15868,
        239,
        11,
        27364,
        2415,
        250,
        11,
        8225,
        13486,
        17047,
        22752,
        11,
        31217,
        13127,
        17047,
        10251,
        11,
        10960,
        6180,
        7182,
        33744,
        232,
        21670,
        1543,
        6723,
        4622,
        3581,
        17238,
        238,
        16516,
        29649,
        13127,
        5708,
        36979,
        11,
        12467,
        39560,
        2289,
        8990,
        1960,
        16657,
        1546,
        10960,
        29389,
        1543,
        35181,
        5620,
        4715,
        40,
        1546,
        32943,
        9254,
        26748,
        42716,
        2289,
        2131,
        23575,
        35477,
        11,
        31929,
        9249,
        4913,
        13545,
        34688,
        39806,
        23987,
        1546,
        32943,
        9254,
        30622,
        36841,
        11,
        8395,
        6723,
        21068,
        11284,
        8661,
        39156,
        32316,
        1486,
        114,
        45569,
        34,
        1429,
        5708,
        6734,
        33597,
        19497,
        1543,
        51365
      ],
      "temperature": 0,
      "avg_logprob": -0.15367189134870257,
      "compression_ratio": 1.1928783382789319,
      "no_speech_prob": 1.8638830062700507e-11,
      "source": "whisper"
    },
    {
      "id": 4,
      "seek": 9332,
      "start": 93.32,
      "end": 123.32,
      "text": "OpenCode不光是個agent工具,它還提供這種大模型的調用的服務,裡面的DeepSeek就是官方提供的。套餐會有一些折扣,所以比DeepSeek官方去安量收費要便宜一點。而且它又是個開源的項目,所以它不限制去接什麼其他的工具。所以很多人,包括我在內,都從OpenCode裡面去使用DeepSeek。OpenCode裡面提供免費的V4Flash,付費套餐Go裡面也有收費版。我自己訂了這個套餐,實際用下來,免費版跟收費版其實沒有太大差別,只是免費版它有調用次數的限制。然後接著我們來看具體的數字。OpenCode的工具,",
      "tokens": [
        50365,
        45569,
        34,
        1429,
        1960,
        20690,
        1541,
        3338,
        559,
        317,
        23323,
        39806,
        11,
        11284,
        7824,
        20949,
        3254,
        249,
        33005,
        3582,
        41908,
        39823,
        1546,
        32943,
        9254,
        1546,
        27408,
        39517,
        11,
        32399,
        1546,
        11089,
        595,
        10637,
        916,
        5620,
        31929,
        9249,
        20949,
        3254,
        249,
        1546,
        1543,
        1881,
        245,
        22217,
        238,
        6236,
        32241,
        13824,
        7235,
        246,
        3416,
        96,
        11,
        7239,
        11706,
        11089,
        595,
        10637,
        916,
        31929,
        9249,
        6734,
        16206,
        26748,
        18681,
        50071,
        4275,
        27364,
        2415,
        250,
        19202,
        1543,
        22942,
        11284,
        17047,
        1541,
        3338,
        8949,
        47402,
        1546,
        8313,
        227,
        11386,
        11,
        7239,
        11284,
        1960,
        43446,
        25491,
        6734,
        14468,
        7598,
        9572,
        31309,
        23323,
        39806,
        1543,
        7239,
        20778,
        4035,
        11,
        41828,
        40528,
        28472,
        11,
        7182,
        21068,
        45569,
        34,
        1429,
        32399,
        6734,
        22982,
        9254,
        11089,
        595,
        10637,
        916,
        1543,
        45569,
        34,
        1429,
        32399,
        20949,
        3254,
        249,
        2347,
        235,
        50071,
        1546,
        53,
        19,
        25680,
        1299,
        11,
        41469,
        50071,
        1881,
        245,
        22217,
        238,
        12104,
        32399,
        6404,
        2412,
        18681,
        50071,
        42096,
        1543,
        1654,
        17645,
        28320,
        2289,
        6287,
        1881,
        245,
        22217,
        238,
        11,
        10376,
        34837,
        9254,
        27769,
        11,
        2347,
        235,
        50071,
        42096,
        9678,
        18681,
        50071,
        42096,
        14139,
        6963,
        9455,
        3582,
        21679,
        16158,
        11,
        36859,
        2347,
        235,
        50071,
        42096,
        11284,
        2412,
        32943,
        9254,
        9487,
        30622,
        1546,
        43446,
        25491,
        1543,
        10213,
        14468,
        19382,
        5884,
        3763,
        4200,
        39806,
        23987,
        1546,
        30622,
        22381,
        1543,
        45569,
        34,
        1429,
        1546,
        23323,
        39806,
        11
      ],
      "temperature": 0,
      "avg_logprob": -0.1147877816800718,
      "compression_ratio": 1.4444444444444444,
      "no_speech_prob": 1.671042818007784e-11,
      "source": "whisper"
    },
    {
      "id": 5,
      "seek": 12332,
      "start": 123.32,
      "end": 153.29999999999998,
      "text": "官方說,8月1日,V4Flash發布的第二天,免費版燒了5T的Token,收費版燒了3T的Token,第二天就是8T。這還只是開始,OpenCode有一個實時的數據頁面,收費Token的日消耗量。8月5日已經漲到了6.3T,是Flash正式版發布前的4倍。如果我們再去按照免費跟收費版的比例來推算一下,免費部分一天還要再燒10.5T左右的Token,加起來一天大概要燒17T左右的Token。17T是個什麼概念?就是OpenRoot,我們知道它是全球最大的模型的主軟,",
      "tokens": [
        50365,
        31929,
        9249,
        4622,
        11,
        23,
        6939,
        16,
        6890,
        11,
        53,
        19,
        25680,
        1299,
        14637,
        34688,
        1546,
        19693,
        6135,
        11,
        2347,
        235,
        50071,
        42096,
        24184,
        240,
        2289,
        20,
        51,
        1546,
        51,
        8406,
        11,
        18681,
        50071,
        42096,
        24184,
        240,
        2289,
        18,
        51,
        1546,
        51,
        8406,
        11,
        19693,
        6135,
        5620,
        23,
        51,
        1543,
        2664,
        7824,
        36859,
        21017,
        11,
        45569,
        34,
        1429,
        2412,
        8990,
        10376,
        6611,
        1546,
        30622,
        36841,
        8313,
        223,
        8833,
        11,
        18681,
        50071,
        51,
        8406,
        1546,
        6890,
        28837,
        4450,
        245,
        26748,
        1543,
        23,
        6939,
        20,
        6890,
        25338,
        14065,
        110,
        21381,
        21,
        13,
        18,
        51,
        11,
        1541,
        25680,
        1299,
        15789,
        27584,
        42096,
        14637,
        34688,
        8945,
        1546,
        19,
        35477,
        1543,
        13119,
        5884,
        8623,
        6734,
        26613,
        32150,
        2347,
        235,
        50071,
        9678,
        18681,
        50071,
        42096,
        1546,
        11706,
        17797,
        3763,
        33597,
        19497,
        8861,
        11,
        2347,
        235,
        50071,
        32174,
        2257,
        6135,
        7824,
        4275,
        8623,
        24184,
        240,
        3279,
        13,
        20,
        51,
        29457,
        1546,
        51,
        8406,
        11,
        9990,
        21670,
        2257,
        6135,
        32044,
        4275,
        24184,
        240,
        7773,
        51,
        29457,
        1546,
        51,
        8406,
        1543,
        7773,
        51,
        1541,
        3338,
        7598,
        26181,
        33679,
        30,
        5620,
        45569,
        49,
        6259,
        11,
        5884,
        7758,
        11284,
        1541,
        11319,
        28533,
        8661,
        39156,
        41908,
        39823,
        1546,
        13557,
        14454,
        253,
        11,
        51864
      ],
      "temperature": 0,
      "avg_logprob": -0.10840900203724599,
      "compression_ratio": 1.3489583333333333,
      "no_speech_prob": 1.5218291904428582e-11,
      "source": "whisper"
    },
    {
      "id": 6,
      "seek": 15332,
      "start": 153.32,
      "end": 182.82,
      "text": "基本上沒有質疑,幾百個大模型,800萬的用戶,全平台加起來一天也就3T多的Token。OpenCode在DeepSeek一個模型上就要燒的量,是整個OpenRooter幾百個模型用量的總和的5倍。所以我們來推測一下,官方沒有說漲價的原因,但我認為最有可能的解釋就是,算力真的不太夠了。於是Flash變強之後,吸引來的是Agent的重度用戶,工具一接上之後,Token就是沉吊的燒起來了。光是OpenCode這一個客戶一天就能干到17T,用量一周直接漲了4倍。",
      "tokens": [
        50365,
        37946,
        5708,
        6963,
        43450,
        18140,
        239,
        11,
        23575,
        31906,
        3338,
        3582,
        41908,
        39823,
        11,
        14423,
        14099,
        1546,
        9254,
        1486,
        114,
        11,
        11319,
        16716,
        15433,
        9990,
        21670,
        2257,
        6135,
        6404,
        3111,
        18,
        51,
        6392,
        1546,
        51,
        8406,
        1543,
        45569,
        34,
        1429,
        3581,
        11089,
        595,
        10637,
        916,
        8990,
        41908,
        39823,
        5708,
        3111,
        4275,
        24184,
        240,
        1546,
        26748,
        11,
        1541,
        27662,
        3338,
        45569,
        49,
        6259,
        260,
        23575,
        31906,
        3338,
        41908,
        39823,
        9254,
        26748,
        1546,
        26575,
        12565,
        1546,
        20,
        35477,
        1543,
        7239,
        5884,
        3763,
        33597,
        9592,
        105,
        8861,
        11,
        31929,
        9249,
        6963,
        4622,
        14065,
        110,
        45441,
        1546,
        42577,
        11,
        8395,
        1654,
        39450,
        8661,
        2412,
        16657,
        1546,
        17278,
        5873,
        233,
        5620,
        11,
        19497,
        13486,
        8034,
        1960,
        9455,
        31649,
        2289,
        1543,
        19488,
        1541,
        25680,
        1299,
        26327,
        22752,
        26362,
        11,
        46431,
        41301,
        3763,
        24620,
        32,
        6930,
        1546,
        12624,
        13127,
        9254,
        1486,
        114,
        11,
        23323,
        39806,
        2257,
        14468,
        5708,
        26362,
        11,
        51,
        8406,
        5620,
        3308,
        231,
        2392,
        232,
        1546,
        24184,
        240,
        21670,
        2289,
        1543,
        20690,
        1541,
        45569,
        34,
        1429,
        2664,
        8990,
        32316,
        1486,
        114,
        2257,
        6135,
        3111,
        8225,
        26111,
        4511,
        7773,
        51,
        11,
        9254,
        26748,
        2257,
        37390,
        43297,
        14065,
        110,
        2289,
        19,
        35477,
        1543,
        51840
      ],
      "temperature": 0,
      "avg_logprob": -0.13111423693205182,
      "compression_ratio": 1.3325062034739454,
      "no_speech_prob": 1.440752205039697e-11,
      "source": "whisper"
    },
    {
      "id": 7,
      "seek": 18282,
      "start": 182.82,
      "end": 194.82,
      "text": "但是背後這個卡這個算力不會一周之內就多出4倍,能立刻的去見效。所以只能拿價格去來做暫時的調節。好了,接下來趕緊跟大家來分享一下快訊。然後,今天視頻就到這裡,我是Dizon黑心李少,我們下次見。",
      "tokens": [
        50365,
        11189,
        46329,
        5661,
        6287,
        32681,
        6287,
        19497,
        13486,
        21121,
        2257,
        37390,
        9574,
        28472,
        3111,
        6392,
        7781,
        19,
        35477,
        11,
        8225,
        24409,
        45500,
        1546,
        6734,
        10470,
        43076,
        1543,
        7239,
        14003,
        8225,
        24351,
        45441,
        30921,
        6734,
        3763,
        10907,
        23826,
        104,
        6611,
        1546,
        32943,
        27694,
        1543,
        12621,
        11,
        40012,
        9759,
        243,
        30172,
        47349,
        3763,
        30855,
        8861,
        10251,
        5396,
        232,
        1543,
        10213,
        11,
        12074,
        27333,
        29940,
        45918,
        18771,
        11,
        15914,
        35,
        590,
        266,
        32141,
        7945,
        31206,
        15686,
        11,
        5884,
        35829,
        10470,
        1543,
        50965
      ],
      "temperature": 0,
      "avg_logprob": -0.3137282859988329,
      "compression_ratio": 1.146551724137931,
      "no_speech_prob": 1.6435755534338625e-11,
      "source": "whisper"
    },
    {
      "id": 8,
      "seek": 19482,
      "start": 194.82,
      "end": 196.82,
      "text": "謝謝。",
      "tokens": [
        50365,
        11003,
        1543,
        50465
      ],
      "temperature": 0,
      "avg_logprob": -0.9122788429260253,
      "compression_ratio": 0.5,
      "no_speech_prob": 1.8138837654113615e-10,
      "source": "whisper"
    }
  ],
  "transcript_selection": {
    "schema": "videonote.transcript_selection.v1",
    "status": "SELECTED",
    "requested_language": "zh",
    "duration": 195.1405,
    "reference_path": null,
    "reference_kind": null,
    "reference_language": null,
    "selected_source": "whisper_segmented",
    "decision_reason": "source_unavailable_or_untrusted_whisper_passed",
    "whisper_rejected": false,
    "source_profile": {
      "segment_count": 0,
      "duration_coverage": 0.0,
      "usable_full": false,
      "usable_partial": false,
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "language": {
        "requested": "zh",
        "matched": false
      }
    },
    "whisper_profile": {
      "segment_count": 9,
      "first_start": 0.0,
      "final_end": 196.82,
      "duration_coverage": 1.0,
      "maximum_gap_seconds": 0.02,
      "inverted_count": 0,
      "language": {
        "requested": "zh",
        "matched": true,
        "latin": 300,
        "han": 1073,
        "kana": 0,
        "hangul": 0
      },
      "hallucination_region_count": 0,
      "hallucination_regions": [],
      "usable_full": true,
      "usable_partial": true
    },
    "comparison": {
      "window_seconds": 30.0,
      "compared_window_count": 0,
      "median_score": 0.0,
      "divergent_window_count": 0,
      "divergent_ratio": 0.0,
      "divergent_windows": []
    }
  }
}