{
  "model": "Nox",
  "release": "v1.3.1",
  "release_binding": {
    "repo_id": "llm-semantic-router/Decision-1.0-Nox",
    "revision": "74979f7c7408325716dcb89b686d9cf94301923a",
    "release_tag": "v1.3.1",
    "bundle_manifest_sha256": "83876db506b2d98e3e8ce7d34310b21f97bac30d4f5aef3371053798bff08830",
    "publication_receipt_sha256": "7d680d3a1dc5919fc2975cbea9ea5650f39cdbf317d8e1ca6b2d03d5ae11bcdb"
  },
  "condition": "distinct_primary",
  "type": "choice",
  "Q": [
    1,
    2,
    4,
    8,
    16,
    32
  ],
  "tokens_per_question": 499,
  "samples_per_point": 30,
  "fresh_process_blocks": 6,
  "points": {
    "1": {
      "count": 30,
      "p50_ms": 32.678454,
      "p95_ms": 33.71943,
      "mean_ms": 32.7633463,
      "std_ms": 0.5253808729894747,
      "minimum_ms": 31.98487,
      "maximum_ms": 33.745036,
      "questions_per_second_at_p50": 30.601202859841532,
      "peak_allocated_bytes": 8675011072,
      "peak_reserved_bytes": 9246343168,
      "questions": 1,
      "tokens_per_question": 499
    },
    "2": {
      "count": 30,
      "p50_ms": 36.370171,
      "p95_ms": 36.64674145,
      "mean_ms": 36.389989299999996,
      "std_ms": 0.11128765575847639,
      "minimum_ms": 36.268881,
      "maximum_ms": 36.692332,
      "questions_per_second_at_p50": 54.99011813829525,
      "peak_allocated_bytes": 8746551808,
      "peak_reserved_bytes": 9246343168,
      "questions": 2,
      "tokens_per_question": 499
    },
    "4": {
      "count": 30,
      "p50_ms": 60.415027,
      "p95_ms": 60.9524315,
      "mean_ms": 60.47764396666667,
      "std_ms": 0.286383903775007,
      "minimum_ms": 60.134886,
      "maximum_ms": 61.481055,
      "questions_per_second_at_p50": 66.20869341000211,
      "peak_allocated_bytes": 8890681856,
      "peak_reserved_bytes": 9246343168,
      "questions": 4,
      "tokens_per_question": 499
    },
    "8": {
      "count": 30,
      "p50_ms": 103.619237,
      "p95_ms": 104.1805109,
      "mean_ms": 103.640764,
      "std_ms": 0.34493524425007943,
      "minimum_ms": 102.972861,
      "maximum_ms": 104.235019,
      "questions_per_second_at_p50": 77.20574124667604,
      "peak_allocated_bytes": 9178941952,
      "peak_reserved_bytes": 9246343168,
      "questions": 8,
      "tokens_per_question": 499
    },
    "16": {
      "count": 30,
      "p50_ms": 206.9910775,
      "p95_ms": 207.62369065000001,
      "mean_ms": 206.9661473,
      "std_ms": 0.49304760661506103,
      "minimum_ms": 206.100008,
      "maximum_ms": 207.740005,
      "questions_per_second_at_p50": 77.2980178336431,
      "peak_allocated_bytes": 9178946560,
      "peak_reserved_bytes": 9246343168,
      "questions": 16,
      "tokens_per_question": 499
    },
    "32": {
      "count": 30,
      "p50_ms": 414.19944399999997,
      "p95_ms": 415.3707452,
      "mean_ms": 414.2741519666667,
      "std_ms": 0.6025438072196239,
      "minimum_ms": 413.162275,
      "maximum_ms": 416.066377,
      "questions_per_second_at_p50": 77.25746729877311,
      "peak_allocated_bytes": 9178946560,
      "peak_reserved_bytes": 9246343168,
      "questions": 32,
      "tokens_per_question": 499
    }
  },
  "timing_summary_sha256": "501f25efd1813b566d77e98278229440b7a3b4a2358def1462c0171a33e9f3be",
  "API_sha256": "273f6f10f22d5a68b8db34cfcbd35407fb43d8030f6d7cd188bcf118dd90a152",
  "scope": "Only optimized path subsequently published, no old-release series. Fixed-length Python requests, excludes network and loading.",
  "measurement_applicability": {
    "measured_API_sha256": "273f6f10f22d5a68b8db34cfcbd35407fb43d8030f6d7cd188bcf118dd90a152",
    "current_API_sha256": "6716bdabca3cd2f1aa447d42a6cf53ee62d7ea97984a01f16b4c09fac8aaf17e",
    "measured_bundle_manifest_sha256": "83876db506b2d98e3e8ce7d34310b21f97bac30d4f5aef3371053798bff08830",
    "current_bundle_manifest_sha256": "7d9b06bc25a75b1f131aabd280df0ef2777bf69db98574a300067ace40c61c9c",
    "current_API_byte_identical_to_measured": false,
    "measured_criteria_all_have_non_null_descriptions": true,
    "runtime_change": "Only Choice description=None is replaced with that option key; explicit descriptions use the same rendering path.",
    "same_measured_points_retained": true,
    "current_null_description_latency_not_measured_by_this_curve": true,
    "no_HTTP_throughput_claim": true
  }
}