WORKLOAD REPORT · 2.0.0

Run 3: matched retained evidence, benchmark-window modeled cost only

2026-09-24T06:22:30.638Z

Imported producer results and user-supplied commercial inputs. No independent hardware rerun or source authenticity is claimed.

COMPARISON_OF_SUPPLIED_EVIDENCE 33.1% lower modeled cost for A relative to B.

Hot Aisle

Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8

$0.6908

per 1,000 evaluator-passed, latency-qualified requests

Result rate
1.202 / second
Completed / attempted
8,622 / 8,622
Trials / total measured seconds
1 / 3,606.58
Allocation
1 GPUs × $2.99/hour + $0.0000/hour extras
Cost basis
modeled charge for summed benchmark windows; $3.00
Price source / period
campaign listed price; benchmark window model only / 2026-09-24
Quote / cost note
None

Successful-request latency (ms)

p50p95p99
TTFT52.898155.038568.93
pooled successful-request samples
E2E2,280.24112,353.63915,339.371
pooled successful-request samples
Conditions and input commitments
{
  "identity": {
    "model": "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8",
    "revision": "dcaee4d4dfc5ee71ad501f01f530e5652438fde0",
    "precision": "FP8",
    "tokenizer": "dcaee4d4dfc5ee71ad501f01f530e5652438fde0",
    "workload": "sha256:6c508237082261ac927b96825e2f097782389b1cebc3f78360f0665a8db3501f",
    "cache": "no-prefix-cache",
    "load": "azure-code:54e9a6d2a4bd06ba1e060304b900abbc74cbea53de96506e60fe5bb4f2277fb6:2023-11-16T18:17:04Z:factor=0.95:duration=3600:seed=700000:workers=256:max_tokens=1024:temperature=0.2",
    "runtime": "vllm/vllm-openai-rocm:v0.30.0@sha256:2e7da1ad1c66836802072588adea75f9f4991da5f9545b4318e91d422c22ce6a"
  },
  "gates": {
    "ttft_ms": 1000,
    "e2e_ms": 60000,
    "include_client_queue": true,
    "quality": true
  },
  "criterion": "evalplus-0.3.1-human-v0.1.10-mbpp-v0.2.0-base-and-plus-raw-completion-v1",
  "holds": [],
  "warnings": []
}

Input commitments

SHA-256 ef907e4cbe45832890109c4ca0d37851ad8e4f087beb70421506af897e6a67ab
Record 0 · 605631 bytes

DigitalOcean

Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8

$1.03

per 1,000 evaluator-passed, latency-qualified requests

Result rate
1.187 / second
Completed / attempted
8,622 / 8,622
Trials / total measured seconds
1 / 3,606.393
Allocation
1 GPUs × $4.41/hour + $0.0000/hour extras
Cost basis
modeled charge for summed benchmark windows; $4.42
Price source / period
campaign listed price; benchmark window model only / 2026-09-24
Quote / cost note
None

Successful-request latency (ms)

p50p95p99
TTFT34.70376.1972,233.607
pooled successful-request samples
E2E2,298.88812,067.81215,998.755
pooled successful-request samples
Conditions and input commitments
{
  "identity": {
    "model": "Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8",
    "revision": "dcaee4d4dfc5ee71ad501f01f530e5652438fde0",
    "precision": "FP8",
    "tokenizer": "dcaee4d4dfc5ee71ad501f01f530e5652438fde0",
    "workload": "sha256:6c508237082261ac927b96825e2f097782389b1cebc3f78360f0665a8db3501f",
    "cache": "no-prefix-cache",
    "load": "azure-code:54e9a6d2a4bd06ba1e060304b900abbc74cbea53de96506e60fe5bb4f2277fb6:2023-11-16T18:17:04Z:factor=0.95:duration=3600:seed=700000:workers=256:max_tokens=1024:temperature=0.2",
    "runtime": "vllm/vllm-openai@sha256:8a69ffad015f138d7170c4ddc429e230a3bc1c1719f67e14324749df200a4b90"
  },
  "gates": {
    "ttft_ms": 1000,
    "e2e_ms": 60000,
    "include_client_queue": true,
    "quality": true
  },
  "criterion": "evalplus-0.3.1-human-v0.1.10-mbpp-v0.2.0-base-and-plus-raw-completion-v1",
  "holds": [],
  "warnings": []
}

Input commitments

SHA-256 f1a435b029675ffd2a7c6bb96fc0526c5294bb8acdda9ac76bbe764f9fb99b77
Record 0 · 612180 bytes