32 GB · 53 reports

vGPU-32GB · 32 GB

Qualified results first; original reports and full tables stay below.

Decision · Qualified results

Speeds compare only within one exact test. Recipe availability is checked on each report against the published recipe executor.

clef-flash
ms:Cloudflare · 4 tested configurations
- bf16 · Text → Structured data · typed-invoices-v1 · Test duration 6 s
  Throughput · Batch 1
  10.6decisions/sec
  Best measured · Batch 4
  11decisions/sec
  Observed GPU occupancy
  18.2GiB
  P95 latency
  99 ms
  Use this recipe · Batch 4 ↗Evidence report and run availability ↗Same test on other GPUs
  Test details · 2 batch sizes
  Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
  Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
  Engine
  clef-native · PyTorch 2.13.0+cu130
  Protocol
  native-forward-v1
  Dataset
  typed-invoices-v1
  Workload
  decision
  Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
  B1 10.56 99 ms — 18.17 206.1 2 63f44036 ↗
  B4 11.04 402 ms — 18.55 244.4 2 63f44036 ↗
- 3 other test configurations on this GPU
  - bf16 · Text → Structured data · typed-invoices-v1 · Test duration 6 s
    Throughput · Batch 1
    10.6decisions/sec
    Best measured · Batch 4
    11.1decisions/sec
    Observed GPU occupancy
    18.2GiB
    P95 latency
    103 ms
    Use this recipe · Batch 4 ↗Evidence report and run availability ↗Same test on other GPUs
    Test details · 2 batch sizes
    Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
    Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
    Engine
    clef-native
    Protocol
    native-forward-v1
    Dataset
    typed-invoices-v1
    Workload
    decision
    Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
    B1 10.56 103 ms — 18.17 196.8 1 4eb73a4d ↗
    B4 11.1 390 ms — 18.56 235.9 1 4eb73a4d ↗
  - bf16 · Text → Structured data · typed-invoices-v1 · Test duration 1 s
    Throughput · Batch 1
    10.3decisions/sec
    Best measured · Batch 4
    10.9decisions/sec
    Observed GPU occupancy
    18.2GiB
    P95 latency
    104 ms
    Use this recipe · Batch 4 ↗Evidence report and run availability ↗Same test on other GPUs
    Test details · 2 batch sizes
    Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
    Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
    Engine
    clef-native
    Protocol
    native-forward-v1
    Dataset
    typed-invoices-v1
    Workload
    decision
    Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
    B1 10.26 104 ms — 18.17 67.9 1 1d1661ad ↗
    B4 10.89 389 ms — 18.49 156 1 1d1661ad ↗
  - bf16 · 1K → 1K tokens
    Throughput · Batch 1
    10.2decisions/sec
    Best measured · Batch 4
    10.9decisions/sec
    Observed GPU occupancy
    18.2GiB
    P95 latency
    104 ms
    Use this recipe · Batch 4 ↗Evidence report and run availability ↗Same test on other GPUs
    Test details · 2 batch sizes
    Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.
    Reports combined counts reports, not inference requests. Request successes and failures are listed in each report.
    Engine
    vLLM
    Protocol
    native-forward-v1
    Dataset
    typed-invoices-v1
    Workload
    decision
    Batch size Mean throughput (decisions/sec) P95 TTFT Observed occupancy (GiB) GPU power (W) Reports combined View report
    B1 10.19 104 ms — 18.17 69.8 1 caef7154 ↗
    B4 10.88 388 ms — 18.49 152.8 1 caef7154 ↗

Batch size	Mean throughput (decisions/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	10.56	99 ms	—	18.17	206.1	2	63f44036 ↗
B4	11.04	402 ms	—	18.55	244.4	2	63f44036 ↗

Batch size	Mean throughput (decisions/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	10.56	103 ms	—	18.17	196.8	1	4eb73a4d ↗
B4	11.1	390 ms	—	18.56	235.9	1	4eb73a4d ↗

Batch size	Mean throughput (decisions/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	10.26	104 ms	—	18.17	67.9	1	1d1661ad ↗
B4	10.89	389 ms	—	18.49	156	1	1d1661ad ↗

Batch size	Mean throughput (decisions/sec)	P95	TTFT	Observed occupancy (GiB)	GPU power (W)	Reports combined	View report
B1	10.19	104 ms	—	18.17	69.8	1	caef7154 ↗
B4	10.88	388 ms	—	18.49	152.8	1	caef7154 ↗

Occupancy includes engine buffers and reserved KV cache. It is not the minimum VRAM required by the model.

Sortable tables by exact test condition · 8 rows

C: concurrency · B: batch · —: not measured

Decision · 1K → 1K tokens · Batch 1 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	vLLM	decision1,024 → 1,024 tokensnative-forward-v1	B1	10.19	10.19	—	0.104	18.17	69.8	10.19	0.15	1 ↗

Decision · 1K → 1K tokens · Batch 4 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	vLLM	decision1,024 → 1,024 tokensnative-forward-v1	B4	10.88	10.88	—	0.388	18.49	152.8	10.88	0.07	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 1 s · Batch 1 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B1	10.26	10.26	—	0.104	18.17	67.9	10.26	0.15	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 1 s · Batch 4 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B4	10.89	10.89	—	0.389	18.49	156	10.89	0.07	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 6 s · Batch 1 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B1	10.56	10.65	—	0.099	18.17	206.1	10.48	0.05	2 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 6 s · Batch 1 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B1	10.56	10.56	—	0.103	18.17	196.8	10.56	0.05	1 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 6 s · Batch 4 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B4	11.04	11.07	—	0.402	18.55	244.4	11.02	0.05	2 ↗

Decision · Text → Structured data · typed-invoices-v1 · Test duration 6 s · Batch 4 · decisions/sec · 1 rows
Model	Format	Engine	Workload	Concurrency / batch	Mean throughput	Best measured	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	GPU power (W)	Lowest measured	Throughput / W	Reports combined
ms:Cloudflare/clef-flashScaling curves ↗	bf16	clef-native	decisionnative-forward-v1	B4	11.1	11.1	—	0.39	18.56	235.9	11.1	0.05	1 ↗

Scaling curves and individual runs

One model and test protocol per curve; different workloads are never joined into one line.

ms:Cloudflare/clef-flash · bf16 · clef-native · decision · native-forward-v1

Mean throughput decisions/sec

Measured points only. Select a point to open its source report.

Mean run P95 latency s

Measured points only. Select a point to open its source report.

Mean GPU memory used GB

Measured points only. Select a point to open its source report.

Each point is the mean of recorded runs at that level. A source report may differ from the mean; all runs are listed below.

Runs behind the curve · ms:Cloudflare/clef-flash · decisions/sec · 4 rows
Batch size	Throughput	Decode (tok/s)	Uncached prefill (tok/s)	GPU power (W)	TTFT (ms)	P95 (s)	Observed occupancy (GiB)	Test date	View report
1	10.65 decisions/sec	—	Uncached prefill unavailableView report	208.38	—	0.095	18.17	10/7/2026, 12:51:07 PM	63f44036 ↗
1	10.48 decisions/sec	—	Uncached prefill unavailableView report	203.91	—	0.103	18.17	10/7/2026, 11:11:06 AM	c015746c ↗
4	11.02 decisions/sec	—	Uncached prefill unavailableView report	241.97	—	0.415	18.54	10/7/2026, 12:51:07 PM	63f44036 ↗
4	11.07 decisions/sec	—	Uncached prefill unavailableView report	246.75	—	0.389	18.56	10/7/2026, 11:11:06 AM	c015746c ↗

Original reports

Every report for this task, including quality, diagnostic and legacy runs that do not enter comparisons.

Original reports · 6 rows
Model	Test task	Format	Engine	Workload	Protocol	Quality score	Test date
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/7/2026, 12:59:49 PM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/7/2026, 12:51:07 PM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/7/2026, 11:11:06 AM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/7/2026, 10:22:12 AM
clef-flash ↗	Decision	bf16	clef-native	decision	native-forward-v1	—	10/7/2026, 9:55:46 AM
clef-flash ↗	Decision	bf16	vLLM	decision	native-forward-v1	—	10/7/2026, 9:44:25 AM

Running benchmarks

0Updates automatically

No benchmarks are currently reported as running.

Methodology

Average of error-free runs for the selected recorded configuration. Missing methods and fallback runs are not ranked. Performance index = average / fastest matching hardware × 100. Value index = performance per purchase cost / best comparable value × 100. Quality is separate by dataset and protocol. Prices are dated seller listings, not completed transactions or market averages.

Updated 11/5/2026, 11:06:29 PM