DeepSeek V4 Flash 0423: providers compared

Every endpoint serving this model, side by side. ◆ marks the best value in each column — per model, never fleet-wide.

Current state

	State
GMICloudgmicloud/fp8fp8	Healthy	99.99%◆ (best)	99.92%	99.97%◆ (best)	99.69%	1.51s	5.72s◆ (best)	57 tps	1.0M	$0.091 / $0.182
Novitanovita/fp8fp8	Healthy	99.98%	99.96%	99.95%	99.93%◆ (best)	1.34s	6.56s	62 tps	1.0M	$0.140 / $0.280
DigitalOceandigitalocean	Healthy	99.96%	99.81%	99.82%	99.83%	985ms	14.5s	16 tps	1.0M	$0.098 / $0.196
StreamLakestreamlake/fp8fp8	Healthy	99.96%	98.92%	99.04%	97.67%	1.57s	20.5s	59 tps	1.0M	$0.042 / $0.084
SiliconFlowsiliconflow/fp8fp8	Healthy	99.90%	99.41%	99.73%	98.28%	1.07s	6.59s	33 tps	1.0M	$0.130 / $0.280
Relacerelace/fp4fp4	Healthy	99.90%	99.92%	99.64%	99.73%	547ms◆ (best)	6.23s	93 tps	1.0M	$0.030 / $1.28
AtlasCloudatlas-cloud/fp4fp4	Healthy	99.79%	98.76%	98.47%	99.26%	1.89s	14.3s	55 tps	1.0M	$0.140 / $0.280
Baidubaidu/fp8fp8	Healthy	99.70%	99.57%	89.53%	96.37%	1.04s	9.92s	68 tps	1.0M	$0.140 / $0.280
OpenInferenceopen-inference/fp4fp4	Healthy	99.15%	99.56%	98.63%	98.63%	820ms	8.66s	34 tps	1.0M	$0.039 / $1.66
Parasailparasail/fp8fp8	Healthy	99.06%	99.73%	99.80%	99.72%	907ms	12.3s	41 tps	1.0M	$0.140 / $0.280
Alibabaalibaba/fp8fp8	Healthy	98.92%	98.99%	98.78%	99.10%	852ms	6.76s	96 tps◆ (best)	1.0M	$0.134 / $0.268
Mancer 2mancer/fp8fp8	Degraded	98.04%	96.47%	97.13%	97.91%	1.21s	20.7s	18 tps	1.0M	$0.190 / $0.500
Venicevenice	Degraded	95.86%	98.18%	98.88%	98.39%	1.87s	22.5s	28 tps	1.0M	$0.097 / $0.193
Azureazure/us	Degraded	91.92%	96.78%	97.60%	95.91%	4.03s	49.4s	30 tps	1.0M	$0.210 / $0.560
Cloudflarecloudflare	Degraded	90.49%	98.98%	99.72%	99.72%	5.12s	26.6s	46 tps	384K	$0.440 / $1.32
CoreWeavecoreweave/fp8fp8	No data	—	—	—	—	—	—	—	1.0M	$0.140 / $0.280
Crusoecrusoe/fp8fp8	No data	—	—	—	—	—	—	—	1.0M	$0.140 / $0.280
DeepInfradeepinfra/fp8fp8	No data	—	99.80%	99.54%	99.49%	9.58s	24.8s	9.0 tps	1.0M	$0.090 / $0.180
DeepSeekdeepseek	No data	—	—	—	—	—	—	—	1.0M	$0.220 / $0.660
NextBitnextbit/fp8fp8	No data	—	—	—	99.34%	—	—	—	1.0M	$0.150 / $0.300
OpenInferenceopen-inference/fp8fp8	No data	—	—	95.43%	97.88%	—	—	—	1.0M	$0.002 / $1.31◆ (best)
Phalaphala	No data	—	—	—	99.71%	—	—	—	1.0M	$0.200 / $0.400
Rekareka	No data	—	—	98.07%	98.07%	—	—	—	1.0M	$0.130 / $0.260
Sail Researchsail-research/fp4fp4	No data	—	—	—	—	—	—	—	1.0M	$0.090 / $0.180
Waferwafer	No data	—	—	—	99.43%	—	—	—	1.0M	$0.100 / $0.250

GMICloud99.99%◆ (best)
Healthy24h 99.92%
7d
99.97%◆ (best)
30d
99.69%
p50
1.51s
p99
5.72s◆ (best)
Throughput
57 tps
Context
1.0M
$/Mtok
$0.091 / $0.182
gmicloud/fp8fp8
Novita99.98%
Healthy24h 99.96%
7d
99.95%
30d
99.93%◆ (best)
p50
1.34s
p99
6.56s
Throughput
62 tps
Context
1.0M
$/Mtok
$0.140 / $0.280
novita/fp8fp8
DigitalOcean99.96%
Healthy24h 99.81%
7d
99.82%
30d
99.83%
p50
985ms
p99
14.5s
Throughput
16 tps
Context
1.0M
$/Mtok
$0.098 / $0.196
digitalocean
StreamLake99.96%
Healthy24h 98.92%
7d
99.04%
30d
97.67%
p50
1.57s
p99
20.5s
Throughput
59 tps
Context
1.0M
$/Mtok
$0.042 / $0.084
streamlake/fp8fp8
SiliconFlow99.90%
Healthy24h 99.41%
7d
99.73%
30d
98.28%
p50
1.07s
p99
6.59s
Throughput
33 tps
Context
1.0M
$/Mtok
$0.130 / $0.280
siliconflow/fp8fp8
Relace99.90%
Healthy24h 99.92%
7d
99.64%
30d
99.73%
p50
547ms◆ (best)
p99
6.23s
Throughput
93 tps
Context
1.0M
$/Mtok
$0.030 / $1.28
relace/fp4fp4
AtlasCloud99.79%
Healthy24h 98.76%
7d
98.47%
30d
99.26%
p50
1.89s
p99
14.3s
Throughput
55 tps
Context
1.0M
$/Mtok
$0.140 / $0.280
atlas-cloud/fp4fp4
Baidu99.70%
Healthy24h 99.57%
7d
89.53%
30d
96.37%
p50
1.04s
p99
9.92s
Throughput
68 tps
Context
1.0M
$/Mtok
$0.140 / $0.280
baidu/fp8fp8
OpenInference99.15%
Healthy24h 99.56%
7d
98.63%
30d
98.63%
p50
820ms
p99
8.66s
Throughput
34 tps
Context
1.0M
$/Mtok
$0.039 / $1.66
open-inference/fp4fp4
Parasail99.06%
Healthy24h 99.73%
7d
99.80%
30d
99.72%
p50
907ms
p99
12.3s
Throughput
41 tps
Context
1.0M
$/Mtok
$0.140 / $0.280
parasail/fp8fp8
Alibaba98.92%
Healthy24h 98.99%
7d
98.78%
30d
99.10%
p50
852ms
p99
6.76s
Throughput
96 tps◆ (best)
Context
1.0M
$/Mtok
$0.134 / $0.268
alibaba/fp8fp8
Mancer 298.04%
Degraded24h 96.47%
7d
97.13%
30d
97.91%
p50
1.21s
p99
20.7s
Throughput
18 tps
Context
1.0M
$/Mtok
$0.190 / $0.500
mancer/fp8fp8
Venice95.86%
Degraded24h 98.18%
7d
98.88%
30d
98.39%
p50
1.87s
p99
22.5s
Throughput
28 tps
Context
1.0M
$/Mtok
$0.097 / $0.193
venice
Azure91.92%
Degraded24h 96.78%
7d
97.60%
30d
95.91%
p50
4.03s
p99
49.4s
Throughput
30 tps
Context
1.0M
$/Mtok
$0.210 / $0.560
azure/us
Cloudflare90.49%
Degraded24h 98.98%
7d
99.72%
30d
99.72%
p50
5.12s
p99
26.6s
Throughput
46 tps
Context
384K
$/Mtok
$0.440 / $1.32
cloudflare
CoreWeave—
No data24h —
7d
—
30d
—
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.140 / $0.280
coreweave/fp8fp8
Crusoe—
No data24h —
7d
—
30d
—
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.140 / $0.280
crusoe/fp8fp8
DeepInfra—
No data24h 99.80%
7d
99.54%
30d
99.49%
p50
9.58s
p99
24.8s
Throughput
9.0 tps
Context
1.0M
$/Mtok
$0.090 / $0.180
deepinfra/fp8fp8
DeepSeek—
No data24h —
7d
—
30d
—
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.220 / $0.660
deepseek
NextBit—
No data24h —
7d
—
30d
99.34%
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.150 / $0.300
nextbit/fp8fp8
OpenInference—
No data24h —
7d
95.43%
30d
97.88%
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.002 / $1.31◆ (best)
open-inference/fp8fp8
Phala—
No data24h —
7d
—
30d
99.71%
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.200 / $0.400
phala
Reka—
No data24h —
7d
98.07%
30d
98.07%
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.130 / $0.260
reka
Sail Research—
No data24h —
7d
—
30d
—
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.090 / $0.180
sail-research/fp4fp4
Wafer—
No data24h —
7d
—
30d
99.43%
p50
—
p99
—
Throughput
—
Context
1.0M
$/Mtok
$0.100 / $0.250
wafer

Measured 1m ago. Latency and throughput are OpenRouter's rolling 30-minute windows. ◆ marks the best value in each column among these endpoints.

Price vs latency

One dot per endpoint at the latest round · further left = faster, lower = cheaper · hover a dot for its tag

History

Availability

Axis starts at 90% so small differences stay visible; it expands to 0 when an endpoint falls below.

Time to first token

Logarithmic axis — the fleet spans two orders of magnitude. P50 is OpenRouter's rolling 30-minute window, averaged over the hour.

Throughput

Median tokens per second.