{
  "status": "SUPPLEMENT_FROM_EXISTING_SAVED_REQUESTS_NO_NEW_BENCHMARK",
  "source_sha256": "98b50ed08e38a7999d1a9ecb762bea57d813890906c6cde041d6dd334fe5d020",
  "conditions": [
    {
      "model": "flash",
      "workload": "long_document_grounded_qa",
      "concurrency": 1,
      "count": 20,
      "request_sha256": "298f86282e1c7bf2b8298f57a2da0ccddc2524dacd89eabd4f629b212a96a6d2",
      "first_generated": {
        "n": 20,
        "median": 0.33771378255914897,
        "p95": 0.34186441800557077,
        "min": 0.33577759901527315,
        "max": 1.6251776979770511
      },
      "first_final": {
        "n": 20,
        "median": 0.9383142085280269,
        "p95": 1.1295272540301085,
        "min": 0.7098339980002493,
        "max": 2.3320820349035785
      },
      "completion": {
        "n": 20,
        "median": 1.7328429645276628,
        "p95": 1.8890361099038273,
        "min": 1.4095612959936261,
        "max": 3.0232595559209585
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 233.30989206937497,
        "p95": 233.51673737906393,
        "min": 233.20735582848081,
        "max": 233.89188418545348
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 188.18576278761245,
        "p95": 190.55330098821238,
        "min": 108.49217340853924,
        "max": 192.161493418186
      },
      "generated_tokens_including_reasoning": 6348,
      "batch_elapsed_s": 35.17608584591653,
      "batch_throughput_tokens_s": 180.4635122795198,
      "output_tokens": {
        "n": 20,
        "median": 326.0,
        "p95": 347,
        "min": 251,
        "max": 363
      }
    },
    {
      "model": "flash",
      "workload": "scientific_reasoning",
      "concurrency": 1,
      "count": 20,
      "request_sha256": "5cb6e542feac0f1d2fa282d07026b6b848f6456349ebc229f6c75a7ab76cf60f",
      "first_generated": {
        "n": 20,
        "median": 0.06759888294618577,
        "p95": 0.06917750206775963,
        "min": 0.06607994297519326,
        "max": 0.06929693999700248
      },
      "first_final": {
        "n": 20,
        "median": 1.7181373700150289,
        "p95": 20.49449132801965,
        "min": 0.5452149930642918,
        "max": 21.563211795990355
      },
      "completion": {
        "n": 20,
        "median": 2.7944583680364303,
        "p95": 22.859037535032257,
        "min": 1.0758525310084224,
        "max": 23.108468280057423
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 236.49620643573763,
        "p95": 237.73807626868606,
        "min": 234.74014544966838,
        "max": 237.77489650970458
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 231.1710550113903,
        "p95": 234.08684603625193,
        "min": 224.0083961824256,
        "max": 234.11331008333525
      },
      "generated_tokens_including_reasoning": 26566,
      "batch_elapsed_s": 114.0878863359103,
      "batch_throughput_tokens_s": 232.85557172810982,
      "output_tokens": {
        "n": 20,
        "median": 646.0,
        "p95": 5351,
        "min": 241,
        "max": 5410
      }
    },
    {
      "model": "flash",
      "workload": "short_science_qa",
      "concurrency": 1,
      "count": 20,
      "request_sha256": "5e57d8844dbe0bd8383eecb5cd293c02dcf51d5eb3efca5f74bd76eb3574cd30",
      "first_generated": {
        "n": 20,
        "median": 0.06640748347854242,
        "p95": 0.06986434094142169,
        "min": 0.06581807800102979,
        "max": 0.07358841795939952
      },
      "first_final": {
        "n": 20,
        "median": 0.40131706197280437,
        "p95": 0.8619877820601687,
        "min": 0.26341481099370867,
        "max": 0.8620266709476709
      },
      "completion": {
        "n": 20,
        "median": 1.035499827528838,
        "p95": 1.5109517601085827,
        "min": 0.6255984690506011,
        "max": 1.6239429989363998
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 237.79507745544583,
        "p95": 238.12382017903454,
        "min": 237.47287868999558,
        "max": 238.18186706810621
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 223.2754343591733,
        "p95": 227.77293219461794,
        "min": 214.1948975728096,
        "max": 228.45629449000745
      },
      "generated_tokens_including_reasoning": 4916,
      "batch_elapsed_s": 21.94371176394634,
      "batch_throughput_tokens_s": 224.02773299624815,
      "output_tokens": {
        "n": 20,
        "median": 231.0,
        "p95": 344,
        "min": 134,
        "max": 371
      }
    },
    {
      "model": "flash",
      "workload": "long_document_grounded_qa",
      "concurrency": 10,
      "count": 20,
      "request_sha256": "b7359293bf3821a48239f78f197e0571df79544ff692dc616cdab02a1a655b6b",
      "first_generated": {
        "n": 20,
        "median": 0.6747568545397371,
        "p95": 2.6018465739907697,
        "min": 0.36233892501331866,
        "max": 4.123771516024135
      },
      "first_final": {
        "n": 20,
        "median": 4.691072714456823,
        "p95": 5.164662490948103,
        "min": 1.0189679219620302,
        "max": 5.168699292931706
      },
      "completion": {
        "n": 20,
        "median": 5.766351913975086,
        "p95": 9.07819852605462,
        "min": 2.1258481930708513,
        "max": 9.844160073902458
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 70.72263946087935,
        "p95": 138.71060344975893,
        "min": 41.98067108882698,
        "max": 174.65176145019944
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 50.607578220313606,
        "p95": 100.29812302629337,
        "min": 36.35082434614016,
        "max": 145.35374680430039
      },
      "generated_tokens_including_reasoning": 6134,
      "batch_elapsed_s": 11.970924291061237,
      "batch_throughput_tokens_s": 512.4082193536464,
      "output_tokens": {
        "n": 20,
        "median": 306.0,
        "p95": 353,
        "min": 257,
        "max": 377
      }
    },
    {
      "model": "flash",
      "workload": "scientific_reasoning",
      "concurrency": 10,
      "count": 20,
      "request_sha256": "e10232cb3e51eafc905be3c23da8b1fc0dbe2e56cb15af23be587fdecc976dcd",
      "first_generated": {
        "n": 20,
        "median": 0.07481569243827835,
        "p95": 0.14488247490953654,
        "min": 0.0715688019990921,
        "max": 0.14536134607624263
      },
      "first_final": {
        "n": 20,
        "median": 3.119140033959411,
        "p95": 9.893676184001379,
        "min": 0.7429152929689735,
        "max": 20.01637332199607
      },
      "completion": {
        "n": 20,
        "median": 4.737114107992966,
        "p95": 11.747242024051957,
        "min": 1.717243982013315,
        "max": 23.16619716095738
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 155.5353407985436,
        "p95": 189.97029945843573,
        "min": 135.61320978918693,
        "max": 210.87834901824735
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 148.08289711261455,
        "p95": 188.82231145258504,
        "min": 131.85285911071801,
        "max": 210.26325409201075
      },
      "generated_tokens_including_reasoning": 21151,
      "batch_elapsed_s": 26.941651628003456,
      "batch_throughput_tokens_s": 785.0669399204692,
      "output_tokens": {
        "n": 20,
        "median": 679.5,
        "p95": 2135,
        "min": 261,
        "max": 4871
      }
    },
    {
      "model": "flash",
      "workload": "short_science_qa",
      "concurrency": 10,
      "count": 20,
      "request_sha256": "91e98c023827cf0a0c910823c2fbc93c6b030b070ee4fc5f5b643ee402bb3e86",
      "first_generated": {
        "n": 20,
        "median": 0.07753541698912159,
        "p95": 1.4542348550166935,
        "min": 0.07055162906181067,
        "max": 1.4548109759343788
      },
      "first_final": {
        "n": 20,
        "median": 1.4241923029185273,
        "p95": 2.527088513947092,
        "min": 0.45912985398899764,
        "max": 2.574817576096393
      },
      "completion": {
        "n": 20,
        "median": 2.2391338250017725,
        "p95": 3.88343562791124,
        "min": 0.9490037930663675,
        "max": 3.9939204190159217
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 141.09985172793856,
        "p95": 173.68963743266545,
        "min": 78.6842325741302,
        "max": 186.73714798980805
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 100.74113751706582,
        "p95": 168.17413372392554,
        "min": 62.320813380279674,
        "max": 175.50845456734626
      },
      "generated_tokens_including_reasoning": 4728,
      "batch_elapsed_s": 5.3975418109912425,
      "batch_throughput_tokens_s": 875.9543076391134,
      "output_tokens": {
        "n": 20,
        "median": 233.0,
        "p95": 342,
        "min": 144,
        "max": 344
      }
    },
    {
      "model": "flash",
      "workload": "long_document_grounded_qa",
      "concurrency": 20,
      "count": 20,
      "request_sha256": "dd21232a41bc795a30c05409d951c88d313115336f33f05e1fc6109eed108ac3",
      "first_generated": {
        "n": 20,
        "median": 3.0173747719381936,
        "p95": 5.391977772931568,
        "min": 0.3704014589311555,
        "max": 5.660321831936017
      },
      "first_final": {
        "n": 20,
        "median": 6.311283559014555,
        "p95": 6.517397202900611,
        "min": 5.725301862927154,
        "max": 6.771681446000002
      },
      "completion": {
        "n": 20,
        "median": 7.347275770502165,
        "p95": 7.622617239947431,
        "min": 6.783165483968332,
        "max": 7.636848711990751
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 68.65666087571965,
        "p95": 140.7668090890849,
        "min": 38.828810638881485,
        "max": 154.69867117056492
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 40.695873012977344,
        "p95": 44.07953688343619,
        "min": 34.19967733802743,
        "max": 46.90783859944053
      },
      "generated_tokens_including_reasoning": 5906,
      "batch_elapsed_s": 7.638110750936903,
      "batch_throughput_tokens_s": 773.227856021276,
      "output_tokens": {
        "n": 20,
        "median": 296.5,
        "p95": 333,
        "min": 234,
        "max": 349
      }
    },
    {
      "model": "flash",
      "workload": "scientific_reasoning",
      "concurrency": 20,
      "count": 20,
      "request_sha256": "396cd335d1f28aa7cee7f5c6eb54b09b304fac169d923012c57fdcfa5c71a589",
      "first_generated": {
        "n": 20,
        "median": 1.4762819859897718,
        "p95": 1.4775254050036892,
        "min": 0.08681353600695729,
        "max": 1.4780471760313958
      },
      "first_final": {
        "n": 20,
        "median": 4.462011237046681,
        "p95": 22.15388894197531,
        "min": 2.0859039400238544,
        "max": 22.203526912024245
      },
      "completion": {
        "n": 20,
        "median": 5.971965224016458,
        "p95": 23.91087432904169,
        "min": 3.0889872680418193,
        "max": 25.605716536054388
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 153.11101052199237,
        "p95": 196.20168085744942,
        "min": 127.44795342837263,
        "max": 198.87169110458282
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 115.45518343214256,
        "p95": 184.1421580578591,
        "min": 78.66655926815889,
        "max": 187.45814018683333
      },
      "generated_tokens_including_reasoning": 24644,
      "batch_elapsed_s": 25.60904584999662,
      "batch_throughput_tokens_s": 962.3162121834091,
      "output_tokens": {
        "n": 20,
        "median": 689.5,
        "p95": 4403,
        "min": 243,
        "max": 4800
      }
    },
    {
      "model": "flash",
      "workload": "short_science_qa",
      "concurrency": 20,
      "count": 20,
      "request_sha256": "f079ec7f45acbf50b9931c608cc7b4916fb0b96c15e9978ebfed1ccb13b48df0",
      "first_generated": {
        "n": 20,
        "median": 1.4815258929738775,
        "p95": 1.4834380969405174,
        "min": 0.09093543700873852,
        "max": 1.4839434080058709
      },
      "first_final": {
        "n": 20,
        "median": 2.0148734245449305,
        "p95": 2.4973933420842513,
        "min": 1.7942242129938677,
        "max": 2.7658712370321155
      },
      "completion": {
        "n": 20,
        "median": 2.8945703724748455,
        "p95": 3.700675316969864,
        "min": 2.348468965035863,
        "max": 3.745721013052389
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 155.1212055617459,
        "p95": 171.62832471769283,
        "min": 64.93530405452016,
        "max": 172.81075169595817
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 76.0008541786674,
        "p95": 103.2244029213522,
        "min": 57.48425974960177,
        "max": 104.65274873222849
      },
      "generated_tokens_including_reasoning": 4824,
      "batch_elapsed_s": 3.7470324110472575,
      "batch_throughput_tokens_s": 1287.418807955211,
      "output_tokens": {
        "n": 20,
        "median": 220.0,
        "p95": 382,
        "min": 135,
        "max": 392
      }
    },
    {
      "model": "flash",
      "workload": "long_document_grounded_qa",
      "concurrency": 5,
      "count": 20,
      "request_sha256": "b55796fbdd87fd0bba20a097f52daed55b78b02d0eec175ee8d6419651077e07",
      "first_generated": {
        "n": 20,
        "median": 0.35877334902761504,
        "p95": 1.4688742709113285,
        "min": 0.3532568559749052,
        "max": 1.8889481690712273
      },
      "first_final": {
        "n": 20,
        "median": 1.7224093364784494,
        "p95": 3.275654525961727,
        "min": 0.824309010989964,
        "max": 3.4487887570867315
      },
      "completion": {
        "n": 20,
        "median": 3.226141944003757,
        "p95": 5.2848796620965,
        "min": 1.8001638549612835,
        "max": 5.509042778983712
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 112.32090167078641,
        "p95": 174.5865429386167,
        "min": 66.85144001629752,
        "max": 212.34722488173844
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 97.33687742706428,
        "p95": 148.60542844108207,
        "min": 61.23021509089152,
        "max": 171.0955361930785
      },
      "generated_tokens_including_reasoning": 6128,
      "batch_elapsed_s": 14.29237592604477,
      "batch_throughput_tokens_s": 428.76006282713587,
      "output_tokens": {
        "n": 20,
        "median": 305.0,
        "p95": 348,
        "min": 253,
        "max": 402
      }
    },
    {
      "model": "flash",
      "workload": "scientific_reasoning",
      "concurrency": 5,
      "count": 20,
      "request_sha256": "b9e433b2d3821fc5b019a1af36efcbd66bf8c97a3bff42c59f040f8458fe6e38",
      "first_generated": {
        "n": 20,
        "median": 0.07277995650656521,
        "p95": 0.14018646394833922,
        "min": 0.0706230680225417,
        "max": 0.14050113793928176
      },
      "first_final": {
        "n": 20,
        "median": 2.3664547154912725,
        "p95": 17.650894510094076,
        "min": 0.5412435099715367,
        "max": 21.96358710306231
      },
      "completion": {
        "n": 20,
        "median": 3.722883523441851,
        "p95": 19.326786105055362,
        "min": 1.5929365740157664,
        "max": 24.318846564972773
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 181.24391562250827,
        "p95": 197.26593553662156,
        "min": 168.88202641647308,
        "max": 210.17617961459374
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 176.29208457410596,
        "p95": 196.5317711234652,
        "min": 162.22037646462488,
        "max": 209.5905324449629
      },
      "generated_tokens_including_reasoning": 24940,
      "batch_elapsed_s": 36.62509247590788,
      "batch_throughput_tokens_s": 680.9539120319116,
      "output_tokens": {
        "n": 20,
        "median": 642.5,
        "p95": 3701,
        "min": 268,
        "max": 5097
      }
    },
    {
      "model": "flash",
      "workload": "short_science_qa",
      "concurrency": 5,
      "count": 20,
      "request_sha256": "920eb778065aa4364cdfb5d416d969632fe23e7aa443521198bddaff7101ce8e",
      "first_generated": {
        "n": 20,
        "median": 0.07005949248559773,
        "p95": 0.13787962193600833,
        "min": 0.06949802697636187,
        "max": 0.1385479710297659
      },
      "first_final": {
        "n": 20,
        "median": 0.6594311110093258,
        "p95": 1.0045741440262645,
        "min": 0.3152666030218825,
        "max": 1.7766599990427494
      },
      "completion": {
        "n": 20,
        "median": 1.5170406795223244,
        "p95": 2.3339831799967214,
        "min": 0.7997403229819611,
        "max": 2.5184605810791254
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 166.74172113996863,
        "p95": 192.5547320955974,
        "min": 154.6972640824963,
        "max": 207.63011408873018
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 159.61272788020852,
        "p95": 180.9341641586111,
        "min": 149.43056057141007,
        "max": 192.66982227017485
      },
      "generated_tokens_including_reasoning": 4919,
      "batch_elapsed_s": 6.550185419968329,
      "batch_throughput_tokens_s": 750.9711076276348,
      "output_tokens": {
        "n": 20,
        "median": 244.0,
        "p95": 380,
        "min": 135,
        "max": 394
      }
    },
    {
      "model": "glm",
      "workload": "long_document_grounded_qa",
      "concurrency": 1,
      "count": 20,
      "request_sha256": "c30f797089b5e78304cd6c655f89cfec7f98b509c706c3ba734b1c899bb7c31b",
      "first_generated": {
        "n": 20,
        "median": 0.8053059609956108,
        "p95": 0.8112973809475079,
        "min": 0.8031814660644159,
        "max": 0.8179717110469937
      },
      "first_final": {
        "n": 20,
        "median": 0.9779042055015452,
        "p95": 1.1363220389466733,
        "min": 0.8106809130404145,
        "max": 1.137057923944667
      },
      "completion": {
        "n": 20,
        "median": 2.655324879975524,
        "p95": 3.008222009986639,
        "min": 2.250007251976058,
        "max": 3.165838909917511
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 113.53961426163437,
        "p95": 113.84625839134827,
        "min": 113.46195616027445,
        "max": 113.99101605020236
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 79.46259616361746,
        "p95": 83.43799066915105,
        "min": 73.33309697339399,
        "max": 84.96957920294467
      },
      "generated_tokens_including_reasoning": 4203,
      "batch_elapsed_s": 52.972097750985995,
      "batch_throughput_tokens_s": 79.3436578584764,
      "output_tokens": {
        "n": 20,
        "median": 211.0,
        "p95": 251,
        "min": 165,
        "max": 269
      }
    },
    {
      "model": "glm",
      "workload": "scientific_reasoning",
      "concurrency": 1,
      "count": 20,
      "request_sha256": "8edb9038c5945d2fe6a665d91af443771428491468e9b2d320b7492faa20b8f5",
      "first_generated": {
        "n": 20,
        "median": 0.08312511845724657,
        "p95": 0.08875833288766444,
        "min": 0.08098872494883835,
        "max": 0.09098394110333174
      },
      "first_final": {
        "n": 20,
        "median": 6.195475970511325,
        "p95": 61.55289802199695,
        "min": 1.1240539179416373,
        "max": 65.49730760999955
      },
      "completion": {
        "n": 20,
        "median": 9.044314576021861,
        "p95": 68.45510258106515,
        "min": 3.0985547859454527,
        "max": 73.84207955794409
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 119.21675830921168,
        "p95": 120.65244988709057,
        "min": 116.64476457462015,
        "max": 120.75374053772933
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 118.06133983774916,
        "p95": 118.36116709325424,
        "min": 116.51892865840021,
        "max": 118.48564951780509
      },
      "generated_tokens_including_reasoning": 42166,
      "batch_elapsed_s": 359.51973266503774,
      "batch_throughput_tokens_s": 117.28424386453857,
      "output_tokens": {
        "n": 20,
        "median": 1069.5,
        "p95": 7982,
        "min": 365,
        "max": 8604
      }
    },
    {
      "model": "glm",
      "workload": "short_science_qa",
      "concurrency": 1,
      "count": 20,
      "request_sha256": "d59a3acaf4d41e0111dd49f91fad06984db2ad3a589eceac1b44f601c0449ff8",
      "first_generated": {
        "n": 20,
        "median": 0.08947076956974342,
        "p95": 0.09198643802665174,
        "min": 0.0802640609908849,
        "max": 0.09713732497766614
      },
      "first_final": {
        "n": 20,
        "median": 0.0896653970121406,
        "p95": 0.2701649700757116,
        "min": 0.08859507297165692,
        "max": 0.5828283750452101
      },
      "completion": {
        "n": 20,
        "median": 1.3149173440178856,
        "p95": 2.395550201064907,
        "min": 0.9708296019816771,
        "max": 2.6774069279199466
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 122.04239878374081,
        "p95": 122.41683195639803,
        "min": 120.94030630551998,
        "max": 122.48884581857757
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 114.45585055258503,
        "p95": 117.30081877436153,
        "min": 111.20262058198782,
        "max": 117.65114847324335
      },
      "generated_tokens_including_reasoning": 3522,
      "batch_elapsed_s": 30.55408233695198,
      "batch_throughput_tokens_s": 115.27101227126393,
      "output_tokens": {
        "n": 20,
        "median": 150.5,
        "p95": 281,
        "min": 108,
        "max": 315
      }
    },
    {
      "model": "glm",
      "workload": "long_document_grounded_qa",
      "concurrency": 10,
      "count": 20,
      "request_sha256": "006055a18438d8a8dfef7be3d9a0eb0ec18d42196c239529d483439b42e5539e",
      "first_generated": {
        "n": 20,
        "median": 1.6021577029605396,
        "p95": 7.042308849981055,
        "min": 0.8304566210135818,
        "max": 7.701635302975774
      },
      "first_final": {
        "n": 20,
        "median": 4.244003522500861,
        "p95": 7.9441705900244415,
        "min": 1.7590808420209214,
        "max": 8.022276214906015
      },
      "completion": {
        "n": 20,
        "median": 9.919736970507074,
        "p95": 16.609175917925313,
        "min": 3.8395881300093606,
        "max": 16.64883767394349
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 28.525897097009363,
        "p95": 63.09571441490691,
        "min": 18.450287802488834,
        "max": 86.54780911967966
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 20.69850053052221,
        "p95": 48.442695857471186,
        "min": 14.634131391037828,
        "max": 56.28323121055565
      },
      "generated_tokens_including_reasoning": 4288,
      "batch_elapsed_s": 21.038914423901588,
      "batch_throughput_tokens_s": 203.81279725766413,
      "output_tokens": {
        "n": 20,
        "median": 213.0,
        "p95": 268,
        "min": 166,
        "max": 278
      }
    },
    {
      "model": "glm",
      "workload": "scientific_reasoning",
      "concurrency": 10,
      "count": 20,
      "request_sha256": "0ea7f1b80eb0b89def0ee9803c55bc4489b2d90691c4b5e30436c0e6741d0c6c",
      "first_generated": {
        "n": 20,
        "median": 0.10836900846334174,
        "p95": 0.19011063303332776,
        "min": 0.0862355490680784,
        "max": 0.19015290203969926
      },
      "first_final": {
        "n": 20,
        "median": 6.89242602349259,
        "p95": 45.741871702019125,
        "min": 2.011767375981435,
        "max": 62.02254902396817
      },
      "completion": {
        "n": 20,
        "median": 13.064313150534872,
        "p95": 54.63051089295186,
        "min": 4.858911190996878,
        "max": 70.25862582493573
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 77.01421072761987,
        "p95": 95.99598047677229,
        "min": 73.82390506793739,
        "max": 99.59842641176246
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 75.5930006991984,
        "p95": 95.8255728242767,
        "min": 72.66932434555596,
        "max": 99.46109702475655
      },
      "generated_tokens_including_reasoning": 34306,
      "batch_elapsed_s": 81.97168191301171,
      "batch_throughput_tokens_s": 418.5103830906569,
      "output_tokens": {
        "n": 20,
        "median": 964.0,
        "p95": 5235,
        "min": 355,
        "max": 6988
      }
    },
    {
      "model": "glm",
      "workload": "short_science_qa",
      "concurrency": 10,
      "count": 20,
      "request_sha256": "8adbff066bbf39f1589912d3b3be23b00a77211ab50f5a310cc4162ab8c34257",
      "first_generated": {
        "n": 20,
        "median": 0.19312365853693336,
        "p95": 0.2088357510510832,
        "min": 0.10064108902588487,
        "max": 0.20917413593269885
      },
      "first_final": {
        "n": 20,
        "median": 0.20457902742782608,
        "p95": 0.5466623170068488,
        "min": 0.10064108902588487,
        "max": 0.8056584500009194
      },
      "completion": {
        "n": 20,
        "median": 2.4591089515015483,
        "p95": 4.506200293079019,
        "min": 1.344022385077551,
        "max": 4.713204356958158
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 71.03959995845966,
        "p95": 94.81846796593693,
        "min": 64.8641250757462,
        "max": 106.46727689616063
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 66.72445360059898,
        "p95": 92.53916223840778,
        "min": 61.65986000053473,
        "max": 98.83101350540859
      },
      "generated_tokens_including_reasoning": 3609,
      "batch_elapsed_s": 7.2106560110114515,
      "batch_throughput_tokens_s": 500.5092455511214,
      "output_tokens": {
        "n": 20,
        "median": 158.5,
        "p95": 308,
        "min": 97,
        "max": 417
      }
    },
    {
      "model": "glm",
      "workload": "long_document_grounded_qa",
      "concurrency": 20,
      "count": 20,
      "request_sha256": "b71e076189c19bf5181587ebb8806b8b002faa757a95dcf075ceec3426f2e342",
      "first_generated": {
        "n": 20,
        "median": 8.115217046521138,
        "p95": 14.627252261037938,
        "min": 0.9102059700526297,
        "max": 15.419451612979174
      },
      "first_final": {
        "n": 20,
        "median": 12.416893488029018,
        "p95": 15.696549984975718,
        "min": 1.781343883019872,
        "max": 15.707331974990666
      },
      "completion": {
        "n": 20,
        "median": 17.87715112499427,
        "p95": 18.605168865062296,
        "min": 17.168562048929743,
        "max": 19.535411286051385
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 22.500194785563295,
        "p95": 59.801071873183396,
        "min": 12.671777096775443,
        "max": 69.21054889175826
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 11.750087465321137,
        "p95": 13.222132074380196,
        "min": 9.853907985024962,
        "max": 21.55060847378223
      },
      "generated_tokens_including_reasoning": 4385,
      "batch_elapsed_s": 19.53678490302991,
      "batch_throughput_tokens_s": 224.44839423501773,
      "output_tokens": {
        "n": 20,
        "median": 208.5,
        "p95": 246,
        "min": 177,
        "max": 421
      }
    },
    {
      "model": "glm",
      "workload": "scientific_reasoning",
      "concurrency": 20,
      "count": 20,
      "request_sha256": "f6ff55309ae34b2869236a1448822b728f0d99fc94f980ee3ee0c0bc6d6b8b8d",
      "first_generated": {
        "n": 20,
        "median": 0.22311752097448334,
        "p95": 0.22415259899571538,
        "min": 0.0940844650613144,
        "max": 0.22427110595162958
      },
      "first_final": {
        "n": 20,
        "median": 8.467174403485842,
        "p95": 39.24499114998616,
        "min": 1.8707527989754453,
        "max": 67.56364439590834
      },
      "completion": {
        "n": 20,
        "median": 14.061435785493813,
        "p95": 49.213478544028476,
        "min": 4.409369718981907,
        "max": 72.29915442597121
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 69.63135118089754,
        "p95": 89.28085425757118,
        "min": 67.31620189277066,
        "max": 97.8267122098206
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 68.90529677392867,
        "p95": 88.89841013952993,
        "min": 64.40829826027236,
        "max": 97.5391767163848
      },
      "generated_tokens_including_reasoning": 32893,
      "batch_elapsed_s": 72.30101088504307,
      "batch_throughput_tokens_s": 454.94522963584984,
      "output_tokens": {
        "n": 20,
        "median": 969.0,
        "p95": 4375,
        "min": 284,
        "max": 7052
      }
    },
    {
      "model": "glm",
      "workload": "short_science_qa",
      "concurrency": 20,
      "count": 20,
      "request_sha256": "2883a8fae16bd5d1e3769a79916d00835d8105cf7fc6d54ce716f275ae444f94",
      "first_generated": {
        "n": 20,
        "median": 0.23114621749846265,
        "p95": 0.23269876604899764,
        "min": 0.11019432602915913,
        "max": 0.23281141405459493
      },
      "first_final": {
        "n": 20,
        "median": 0.23159599001519382,
        "p95": 0.5334711639443412,
        "min": 0.11019432602915913,
        "max": 1.0730813150294125
      },
      "completion": {
        "n": 20,
        "median": 2.5980550035019405,
        "p95": 4.454065793077461,
        "min": 1.7091614020755515,
        "max": 4.514325515017845
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 71.82095812873726,
        "p95": 84.28988621431337,
        "min": 67.50218231720359,
        "max": 84.76315142881943
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 65.81815158814732,
        "p95": 80.15148778333085,
        "min": 62.018718578173456,
        "max": 80.63220048910478
      },
      "generated_tokens_including_reasoning": 3816,
      "batch_elapsed_s": 4.515418526018038,
      "batch_throughput_tokens_s": 845.1043857866201,
      "output_tokens": {
        "n": 20,
        "median": 171.0,
        "p95": 357,
        "min": 106,
        "max": 364
      }
    },
    {
      "model": "glm",
      "workload": "long_document_grounded_qa",
      "concurrency": 5,
      "count": 20,
      "request_sha256": "d0f338bbc07678987a64fbfdd9fe208fa92a8a2fb3018c1eb83f19fb01644a0e",
      "first_generated": {
        "n": 20,
        "median": 0.8259391925530508,
        "p95": 3.174546162947081,
        "min": 0.8198952539823949,
        "max": 3.8599217699375004
      },
      "first_final": {
        "n": 20,
        "median": 1.7139313379884697,
        "p95": 4.119011331931688,
        "min": 0.8277919030515477,
        "max": 4.291825621970929
      },
      "completion": {
        "n": 20,
        "median": 6.151019945566077,
        "p95": 8.411988867912441,
        "min": 3.1077768630348146,
        "max": 9.668011561967432
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 41.672906881867675,
        "p95": 69.14042462272728,
        "min": 32.4342324727684,
        "max": 100.99511827904996
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 34.90547304546872,
        "p95": 49.868322080737684,
        "min": 26.33330908308707,
        "max": 74.65143420028127
      },
      "generated_tokens_including_reasoning": 4281,
      "batch_elapsed_s": 25.375165123026818,
      "batch_throughput_tokens_s": 168.70826176871597,
      "output_tokens": {
        "n": 20,
        "median": 208.0,
        "p95": 255,
        "min": 166,
        "max": 312
      }
    },
    {
      "model": "glm",
      "workload": "scientific_reasoning",
      "concurrency": 5,
      "count": 20,
      "request_sha256": "3e1b6188c3042914f09758500738d04a6822468890cc77e99fc866687d9bc6dd",
      "first_generated": {
        "n": 20,
        "median": 0.10751061199698597,
        "p95": 1.477903606952168,
        "min": 0.08124216494616121,
        "max": 1.4781383930239826
      },
      "first_final": {
        "n": 20,
        "median": 7.9409920735051855,
        "p95": 59.25610875093844,
        "min": 1.781635291990824,
        "max": 81.92563282395713
      },
      "completion": {
        "n": 20,
        "median": 12.671080571017228,
        "p95": 66.74575074203312,
        "min": 3.986144240014255,
        "max": 88.82718043099158
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 86.9113131943455,
        "p95": 95.24095957298401,
        "min": 77.9786549886578,
        "max": 96.53691045452608
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 85.45174649377614,
        "p95": 95.13979796494213,
        "min": 67.36436343550815,
        "max": 96.39564958774507
      },
      "generated_tokens_including_reasoning": 41784,
      "batch_elapsed_s": 111.38437568303198,
      "batch_throughput_tokens_s": 375.1334039785373,
      "output_tokens": {
        "n": 20,
        "median": 1062.5,
        "p95": 6434,
        "min": 332,
        "max": 8451
      }
    },
    {
      "model": "glm",
      "workload": "short_science_qa",
      "concurrency": 5,
      "count": 20,
      "request_sha256": "e3a6850c3be6ab2970d3c358cadf6d6ef11cca92eef3968b4870bbdd310196a6",
      "first_generated": {
        "n": 20,
        "median": 0.1171531134750694,
        "p95": 0.20452793198637664,
        "min": 0.09455182403326035,
        "max": 0.20472806901670992
      },
      "first_final": {
        "n": 20,
        "median": 0.11723883700324222,
        "p95": 0.3817232290748507,
        "min": 0.09455182403326035,
        "max": 0.5518458139849827
      },
      "completion": {
        "n": 20,
        "median": 2.155462148541119,
        "p95": 3.35147215006873,
        "min": 1.121998441987671,
        "max": 4.97972847300116
      },
      "generation_rate_approx_tokens_s": {
        "n": 20,
        "median": 82.63889120844982,
        "p95": 96.54100389462357,
        "min": 76.41618240669064,
        "max": 106.19342390831923
      },
      "effective_tokens_per_total_wait_s": {
        "n": 20,
        "median": 77.80165788734433,
        "p95": 93.32873003180441,
        "min": 72.75639663550201,
        "max": 98.7577592188295
      },
      "generated_tokens_including_reasoning": 3604,
      "batch_elapsed_s": 10.066758900065906,
      "batch_throughput_tokens_s": 358.0099648533755,
      "output_tokens": {
        "n": 20,
        "median": 165.5,
        "p95": 299,
        "min": 93,
        "max": 384
      }
    }
  ],
  "pairs": [
    {
      "model": "flash",
      "workload": "short_science_qa",
      "comparison": {
        "first_generated": {
          "c1": 0.06640748347854242,
          "c20": 1.4815258929738775,
          "difference_of_medians": 1.415118409495335,
          "ratio_of_medians": 22.309622581204845
        },
        "first_final": {
          "c1": 0.40131706197280437,
          "c20": 2.0148734245449305,
          "difference_of_medians": 1.613556362572126,
          "ratio_of_medians": 5.020652285851406
        },
        "completion": {
          "c1": 1.035499827528838,
          "c20": 2.8945703724748455,
          "difference_of_medians": 1.8590705449460074,
          "ratio_of_medians": 2.795336411965007
        },
        "generation_rate_approx_tokens_s": {
          "c1": 237.79507745544583,
          "c20": 155.1212055617459,
          "difference_of_medians": -82.67387189369992,
          "ratio_of_medians": 0.6523314410947384
        }
      }
    },
    {
      "model": "flash",
      "workload": "long_document_grounded_qa",
      "comparison": {
        "first_generated": {
          "c1": 0.33771378255914897,
          "c20": 3.0173747719381936,
          "difference_of_medians": 2.6796609893790446,
          "ratio_of_medians": 8.934710182903817
        },
        "first_final": {
          "c1": 0.9383142085280269,
          "c20": 6.311283559014555,
          "difference_of_medians": 5.372969350486528,
          "ratio_of_medians": 6.726194169984201
        },
        "completion": {
          "c1": 1.7328429645276628,
          "c20": 7.347275770502165,
          "difference_of_medians": 5.614432805974502,
          "ratio_of_medians": 4.240012465587083
        },
        "generation_rate_approx_tokens_s": {
          "c1": 233.30989206937497,
          "c20": 68.65666087571965,
          "difference_of_medians": -164.6532311936553,
          "ratio_of_medians": 0.2942723956826679
        }
      }
    },
    {
      "model": "flash",
      "workload": "scientific_reasoning",
      "comparison": {
        "first_generated": {
          "c1": 0.06759888294618577,
          "c20": 1.4762819859897718,
          "difference_of_medians": 1.408683103043586,
          "ratio_of_medians": 21.838851792344155
        },
        "first_final": {
          "c1": 1.7181373700150289,
          "c20": 4.462011237046681,
          "difference_of_medians": 2.7438738670316525,
          "ratio_of_medians": 2.597004939720071
        },
        "completion": {
          "c1": 2.7944583680364303,
          "c20": 5.971965224016458,
          "difference_of_medians": 3.1775068559800275,
          "ratio_of_medians": 2.137074322639758
        },
        "generation_rate_approx_tokens_s": {
          "c1": 236.49620643573763,
          "c20": 153.11101052199237,
          "difference_of_medians": -83.38519591374526,
          "ratio_of_medians": 0.6474142347970251
        }
      }
    },
    {
      "model": "glm",
      "workload": "short_science_qa",
      "comparison": {
        "first_generated": {
          "c1": 0.08947076956974342,
          "c20": 0.23114621749846265,
          "difference_of_medians": 0.14167544792871922,
          "ratio_of_medians": 2.583483059439672
        },
        "first_final": {
          "c1": 0.0896653970121406,
          "c20": 0.23159599001519382,
          "difference_of_medians": 0.14193059300305322,
          "ratio_of_medians": 2.5828914802422163
        },
        "completion": {
          "c1": 1.3149173440178856,
          "c20": 2.5980550035019405,
          "difference_of_medians": 1.283137659484055,
          "ratio_of_medians": 1.9758314203714706
        },
        "generation_rate_approx_tokens_s": {
          "c1": 122.04239878374081,
          "c20": 71.82095812873726,
          "difference_of_medians": -50.22144065500355,
          "ratio_of_medians": 0.5884918589317802
        }
      }
    },
    {
      "model": "glm",
      "workload": "long_document_grounded_qa",
      "comparison": {
        "first_generated": {
          "c1": 0.8053059609956108,
          "c20": 8.115217046521138,
          "difference_of_medians": 7.309911085525528,
          "ratio_of_medians": 10.0771848708138
        },
        "first_final": {
          "c1": 0.9779042055015452,
          "c20": 12.416893488029018,
          "difference_of_medians": 11.438989282527473,
          "ratio_of_medians": 12.697453818250706
        },
        "completion": {
          "c1": 2.655324879975524,
          "c20": 17.87715112499427,
          "difference_of_medians": 15.221826245018747,
          "ratio_of_medians": 6.732566421461414
        },
        "generation_rate_approx_tokens_s": {
          "c1": 113.53961426163437,
          "c20": 22.500194785563295,
          "difference_of_medians": -91.03941947607107,
          "ratio_of_medians": 0.19817043533118844
        }
      }
    },
    {
      "model": "glm",
      "workload": "scientific_reasoning",
      "comparison": {
        "first_generated": {
          "c1": 0.08312511845724657,
          "c20": 0.22311752097448334,
          "difference_of_medians": 0.13999240251723677,
          "ratio_of_medians": 2.684116728077068
        },
        "first_final": {
          "c1": 6.195475970511325,
          "c20": 8.467174403485842,
          "difference_of_medians": 2.2716984329745173,
          "ratio_of_medians": 1.366670526007549
        },
        "completion": {
          "c1": 9.044314576021861,
          "c20": 14.061435785493813,
          "difference_of_medians": 5.017121209471952,
          "ratio_of_medians": 1.5547265265157033
        },
        "generation_rate_approx_tokens_s": {
          "c1": 119.21675830921168,
          "c20": 69.63135118089754,
          "difference_of_medians": -49.585407128314145,
          "ratio_of_medians": 0.5840735159086878
        }
      }
    }
  ],
  "definitions": {
    "first_final": "Client request submission to first final-answer content, including server queue, prefill and prior reasoning; not first reasoning chunk.",
    "first_generated": "Client submission to first generated content including reasoning.",
    "generation_rate_approx_tokens_s": "Per request (completion_tokens - 1) / (completion_s - first_generated_s), then median. Includes reasoning and scheduler stalls; client streaming events, not token-level timestamps; approximate, not exact decode speed.",
    "batch_throughput_tokens_s": "Sum of all 20 completion tokens including reasoning / measured condition elapsed seconds. Not per-user speed.",
    "comparison": "Differences and ratios of condition medians, not median paired-request differences. Same task set with variable generated lengths; descriptive single-run comparison."
  },
  "limitations": [
    "20 requests per workload and concurrency; short finite test, not sustained-production capacity or SLA.",
    "1/5/10/20 simultaneous outstanding requests, not people logged in or requests per second.",
    "No per-token timestamps; do not label estimated generation rate exact TPOT or pure GPU decoding throughput.",
    "No separation of queue, input processing and reasoning time in first-final metric.",
    "Different tokenizers; cross-model token/s does not directly measure comparable text volume.",
    "GLM/Flash reasoning settings differ by workload; consult profiles."
  ]
}