{
  "schema": "wasper.parakeet-runtime-configurations.public.v1",
  "comparisonScope": "Observed configurations, not an isolated test of hardware or macOS. App builds and benchmark lifecycle differ.",
  "configurations": [
    {
      "id": "m1-pro",
      "label": "M1 Pro · 16 GB · macOS 14.7.3",
      "snapshot": {
        "schema": "wasper.parakeet-runtime-benchmark.public.v1",
        "version": 1,
        "benchmark": {
          "name": "Local Parakeet runtime comparison",
          "generatedAt": "2026-09-15T00:00:00.000Z",
          "passCount": 3,
          "primaryEvidenceHash": "5b229c5b24408f7d439fb92924c2bfdda5ccb7419e0d134198980ebfdf8fa819"
        },
        "hardware": {
          "model": "MacBook Pro",
          "chip": "Apple M1 Pro",
          "memoryBytes": 17179869184,
          "os": "macOS",
          "osVersion": "14.7.3",
          "powerState": "ac"
        },
        "method": {
          "summary": "Each runtime received the same complete source recording directly on one Mac. Every row runs Parakeet TDT v3 through its local runtime implementation. The Wasper result measures the native Parakeet server, not Wasper.app.",
          "timing": "Response-only timing after one discarded warm-up request.",
          "memory": "The frozen public aggregate does not preserve matching short- and long-recording memory cohorts, so this comparison does not publish sortable physical-memory results.",
          "language": "No language hint or external language-identification service was used."
        },
        "runtimes": [
          {
            "id": "wasper-metal-int8",
            "label": "Wasper Metal INT8",
            "holder": "Wasper",
            "artifact": {
              "kind": "published",
              "label": "osa911/wasper-parakeet-tdt-0.6b-v3-onnx-int8 · v6",
              "url": "https://huggingface.co/osa911/wasper-parakeet-tdt-0.6b-v3-onnx-int8/tree/v6"
            },
            "quantization": "INT8 encoder",
            "sourceRevision": "Wasper 1.8.0",
            "longAudioPolicy": "The native runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.0932977913175933,
                "cer": 0.03953328260318151,
                "realTimeSpeed": 82.7875294722234,
                "medianWallSeconds": 0.11520045799999207
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.2651781026335424,
                "cer": 0.16601886219856665,
                "realTimeSpeed": 139.62638214500413,
                "medianWallSeconds": 5.239361249999998
              }
            }
          },
          {
            "id": "mlx-fp32",
            "label": "MLX Community F32/BF16",
            "holder": "MLX Community",
            "artifact": {
              "kind": "published",
              "label": "mlx-community/parakeet-tdt-0.6b-v3",
              "url": "https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v3"
            },
            "quantization": "F32 weights, BF16 runtime",
            "sourceRevision": "parakeet-mlx 0.5.2 / MLX 0.32.2",
            "longAudioPolicy": "The package windows long recordings internally at 120 seconds with 15 seconds of overlap.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08358720487433359,
                "cer": 0.03134398666521723,
                "realTimeSpeed": 28.83068408517816,
                "medianWallSeconds": 0.3425266249999404
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.3515627158924476,
                "cer": 0.2426173634941756,
                "realTimeSpeed": 33.49278302202962,
                "medianWallSeconds": 21.56856850000005
              }
            }
          },
          {
            "id": "mlx-int8-local",
            "label": "Local MLX INT8",
            "holder": "Wasper",
            "artifact": {
              "kind": "local-derivative",
              "label": "Local group-64 INT8 derivative",
              "baseModelLabel": "mlx-community/parakeet-tdt-0.6b-v3",
              "baseModelUrl": "https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v3"
            },
            "quantization": "INT8 weights, group 64",
            "sourceRevision": "parakeet-mlx 0.5.2 / MLX 0.32.2",
            "longAudioPolicy": "The package windows long recordings internally at 120 seconds with 15 seconds of overlap.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08396801218583397,
                "cer": 0.03076421350146755,
                "realTimeSpeed": 47.23639580761105,
                "medianWallSeconds": 0.21142062500002795
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.3505126150274961,
                "cer": 0.24230805145084858,
                "realTimeSpeed": 53.630992481973976,
                "medianWallSeconds": 13.670069666999392
              }
            }
          },
          {
            "id": "handy-gguf-q8",
            "label": "Handy Q8",
            "holder": "Handy Computer",
            "artifact": {
              "kind": "published",
              "label": "handy-computer/parakeet-tdt-0.6b-v3-gguf Q8_0",
              "url": "https://huggingface.co/handy-computer/parakeet-tdt-0.6b-v3-gguf"
            },
            "quantization": "Q8_0",
            "sourceRevision": "transcribe.cpp 0.2.3 / 63a44d9239",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08625285605483625,
                "cer": 0.03304707033373193,
                "realTimeSpeed": 52.271247,
                "medianWallSeconds": 0.189990875
              }
            },
            "long": {
              "kind": "limited",
              "status": "memory-limit",
              "summary": "Verified through 10m 39s. At 13m 38s, a Metal allocation exceeded the 8 GiB safety policy.",
              "verification": {
                "kind": "verified-through",
                "seconds": 639
              }
            }
          },
          {
            "id": "nvidia-gguf-q8",
            "label": "NVIDIA Q8",
            "holder": "NVIDIA",
            "artifact": {
              "kind": "published",
              "label": "nvidia/parakeet-tdt-0.6b-v3 Q8_0 GGUF",
              "url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3"
            },
            "quantization": "Q8_0",
            "sourceRevision": "NeMo-Speech.cpp 0.1.0 / 4f96762",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08587204874333587,
                "cer": 0.03286589122006015,
                "realTimeSpeed": 47.07873783552122,
                "medianWallSeconds": 0.2120341250002384
              }
            },
            "long": {
              "kind": "limited",
              "status": "empty-output",
              "summary": "The 6m 26s Dutch recording returned empty output in all three passes.",
              "verification": {
                "kind": "not-applicable"
              }
            }
          },
          {
            "id": "istupakov-onnx-int8",
            "label": "Istupakov ONNX INT8",
            "holder": "Istupakov",
            "artifact": {
              "kind": "published",
              "label": "istupakov/parakeet-tdt-0.6b-v3-onnx",
              "url": "https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx"
            },
            "quantization": "INT8",
            "sourceRevision": "onnx-asr 0.12.0 / ONNX Runtime 1.30.0",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.09824828636709824,
                "cer": 0.03844620792115085,
                "realTimeSpeed": 33.864374,
                "medianWallSeconds": 0.293549625
              }
            },
            "long": {
              "kind": "limited",
              "status": "unsupported-input-shape",
              "summary": "A direct long recording exceeds the runtime’s supported tensor shape.",
              "verification": {
                "kind": "not-applicable"
              }
            }
          },
          {
            "id": "fluid-coreml-mixed",
            "label": "Fluid Core ML",
            "holder": "FluidInference",
            "artifact": {
              "kind": "published",
              "label": "FluidInference/parakeet-tdt-0.6b-v3-coreml",
              "url": "https://huggingface.co/FluidInference/parakeet-tdt-0.6b-v3-coreml"
            },
            "quantization": "Mixed precision",
            "sourceRevision": "FluidAudio / 69e42dae8e",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08587204874333587,
                "cer": 0.03206870311990434,
                "realTimeSpeed": 12.756308232604765,
                "medianWallSeconds": 0.7751018330003135
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.20383010473374416,
                "cer": 0.12023484370987697,
                "realTimeSpeed": 35.10841908471022,
                "medianWallSeconds": 22.740842916999945
              }
            }
          }
        ]
      },
      "provenance": "Published September 2026 baseline. Wasper displayed version 1.8.0 but was a locally rebuilt development binary, not necessarily the public 1.8.0 release. The frozen snapshot does not include its exact binary hash."
    },
    {
      "id": "m1-air",
      "label": "M1 Air · 8 GB · macOS 27.0.1",
      "snapshot": {
        "schema": "wasper.parakeet-runtime-benchmark.public.v1",
        "version": 1,
        "benchmark": {
          "name": "Local Parakeet runtime comparison",
          "generatedAt": "2026-10-01T19:05:27.823Z",
          "passCount": 3,
          "primaryEvidenceHash": "b4c25996b2ad568a0d8a0c86b356cfb26ebd3b0e7b5bb8e68c9ebe1c4fa9a4bd"
        },
        "hardware": {
          "model": "MacBook Air",
          "chip": "Apple M1",
          "memoryBytes": 8589934592,
          "os": "macOS",
          "osVersion": "27.0.1",
          "powerState": "ac"
        },
        "method": {
          "summary": "Each runtime received the same complete source recording directly on one Mac. Every row runs Parakeet TDT v3 through its local runtime implementation. The Wasper result measures the native Parakeet server, not Wasper.app.",
          "timing": "Response-only timing after one discarded warm-up request.",
          "memory": "Physical footprint was sampled after each response. These samples are not peak memory or total system memory requirements; no sortable memory ranking is published.",
          "language": "No language hint or external language-identification service was used."
        },
        "runtimes": [
          {
            "id": "wasper-metal-int8",
            "label": "Wasper Metal INT8",
            "holder": "Wasper",
            "artifact": {
              "kind": "published",
              "label": "osa911/wasper-parakeet-tdt-0.6b-v3-onnx-int8 · v6",
              "url": "https://huggingface.co/osa911/wasper-parakeet-tdt-0.6b-v3-onnx-int8/tree/v6"
            },
            "quantization": "INT8 encoder",
            "sourceRevision": "Wasper 1.8.0",
            "longAudioPolicy": "The native runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.09386900228484386,
                "cer": 0.04054788563974345,
                "realTimeSpeed": 41.823288621581874,
                "medianWallSeconds": 0.2295243330001831
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.26501229723381325,
                "cer": 0.1660480425800126,
                "realTimeSpeed": 55.17620476824648,
                "medianWallSeconds": 12.714853000000119
              }
            }
          },
          {
            "id": "mlx-fp32",
            "label": "MLX Community F32/BF16",
            "holder": "MLX Community",
            "artifact": {
              "kind": "published",
              "label": "mlx-community/parakeet-tdt-0.6b-v3",
              "url": "https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v3"
            },
            "quantization": "F32 weights, BF16 runtime",
            "sourceRevision": "parakeet-mlx 0.5.2 / MLX 0.32.2",
            "longAudioPolicy": "The package windows long recordings internally at 120 seconds with 15 seconds of overlap.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08358720487433359,
                "cer": 0.03134398666521723,
                "realTimeSpeed": 16.117507044955527,
                "medianWallSeconds": 0.6121205829996615
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.3515627158924476,
                "cer": 0.2426173634941756,
                "realTimeSpeed": 18.66784115363891,
                "medianWallSeconds": 40.27312249999866
              }
            }
          },
          {
            "id": "mlx-int8-local",
            "label": "Local MLX INT8",
            "holder": "Wasper",
            "artifact": {
              "kind": "local-derivative",
              "label": "Local group-64 INT8 derivative",
              "baseModelLabel": "mlx-community/parakeet-tdt-0.6b-v3",
              "baseModelUrl": "https://huggingface.co/mlx-community/parakeet-tdt-0.6b-v3"
            },
            "quantization": "INT8 weights, group 64",
            "sourceRevision": "parakeet-mlx 0.5.2 / MLX 0.32.2",
            "longAudioPolicy": "The package windows long recordings internally at 120 seconds with 15 seconds of overlap.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08396801218583397,
                "cer": 0.03076421350146755,
                "realTimeSpeed": 26.63498170620817,
                "medianWallSeconds": 0.3778744580000639
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.3505126150274961,
                "cer": 0.24230805145084858,
                "realTimeSpeed": 21.464219475512152,
                "medianWallSeconds": 35.013620207998905
              }
            }
          },
          {
            "id": "handy-gguf-q8",
            "label": "Handy Q8",
            "holder": "Handy Computer",
            "artifact": {
              "kind": "published",
              "label": "handy-computer/parakeet-tdt-0.6b-v3-gguf Q8_0",
              "url": "https://huggingface.co/handy-computer/parakeet-tdt-0.6b-v3-gguf"
            },
            "quantization": "Q8_0",
            "sourceRevision": "transcribe.cpp 0.2.3 / 63a44d9239",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08587204874333587,
                "cer": 0.03286589122006015,
                "realTimeSpeed": 31.672686486560348,
                "medianWallSeconds": 0.3136455420013517
              }
            },
            "long": {
              "kind": "incomplete",
              "completedRequests": 9,
              "expectedRequests": 63,
              "failureCount": 54,
              "summary": "The supplied evidence records runtime errors but does not include detailed failure diagnostics. No memory-limit, timeout, or input-length cause is inferred."
            }
          },
          {
            "id": "nvidia-gguf-q8",
            "label": "NVIDIA Q8",
            "holder": "NVIDIA",
            "artifact": {
              "kind": "published",
              "label": "nvidia/parakeet-tdt-0.6b-v3 Q8_0 GGUF",
              "url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3"
            },
            "quantization": "Q8_0",
            "sourceRevision": "NeMo-Speech.cpp 0.1.0 / 4f96762",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08587204874333587,
                "cer": 0.03290212704279451,
                "realTimeSpeed": 26.45624160436058,
                "medianWallSeconds": 0.3738510410003364
              }
            },
            "long": {
              "kind": "incomplete",
              "completedRequests": 60,
              "expectedRequests": 63,
              "failureCount": 3,
              "summary": "The supplied evidence records runtime errors but does not include detailed failure diagnostics. No memory-limit, timeout, or input-length cause is inferred."
            }
          },
          {
            "id": "istupakov-onnx-int8",
            "label": "Istupakov ONNX INT8",
            "holder": "Istupakov",
            "artifact": {
              "kind": "published",
              "label": "istupakov/parakeet-tdt-0.6b-v3-onnx",
              "url": "https://huggingface.co/istupakov/parakeet-tdt-0.6b-v3-onnx"
            },
            "quantization": "INT8",
            "sourceRevision": "onnx-asr 0.12.0 / ONNX Runtime 1.30.0",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.09824828636709824,
                "cer": 0.03855491538935391,
                "realTimeSpeed": 31.168000793199898,
                "medianWallSeconds": 0.327596958999522
              }
            },
            "long": {
              "kind": "incomplete",
              "completedRequests": 6,
              "expectedRequests": 63,
              "failureCount": 57,
              "summary": "The supplied evidence records runtime errors but does not include detailed failure diagnostics. No memory-limit, timeout, or input-length cause is inferred."
            }
          },
          {
            "id": "fluid-coreml-mixed",
            "label": "Fluid Core ML",
            "holder": "FluidInference",
            "artifact": {
              "kind": "published",
              "label": "FluidInference/parakeet-tdt-0.6b-v3-coreml",
              "url": "https://huggingface.co/FluidInference/parakeet-tdt-0.6b-v3-coreml"
            },
            "quantization": "Mixed precision",
            "sourceRevision": "FluidAudio / 69e42dae8e",
            "longAudioPolicy": "The runtime receives each complete recording directly.",
            "short": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 729,
                "expectedRequests": 729,
                "wer": 0.08625285605483625,
                "cer": 0.031959995651701274,
                "realTimeSpeed": 69.81683166484697,
                "medianWallSeconds": 0.14120150000043213
              }
            },
            "long": {
              "kind": "complete",
              "metrics": {
                "completedRequests": 63,
                "expectedRequests": 63,
                "wer": 0.2077541658606682,
                "cer": 0.12240002801316618,
                "realTimeSpeed": 158.2649313374725,
                "medianWallSeconds": 4.654538917001337
              }
            }
          }
        ]
      },
      "provenance": "October 1, 2026 additional run. Runner 4000ea7 with shutdown timeout locally increased from 5 to 60 seconds. Installed Wasper displayed 1.8.0 and was built locally. The operator confirms the local change was README-only, with no runtime code changes. Native server SHA-256: bdd71e87e1bc032c0b056205a9c55c8e2ca964b76826cfa77525627f3bb9b076. Machine, RAM, macOS marketing version and AC power are reported by the operator; the structured evidence records arm64 and Darwin 27.0.0."
    }
  ]
}
