DeepSeek V3.2: 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
Baidubaidu/fp8fp8	Healthy	100%◆ (best)	99.91%	99.94%	99.94%	999ms	3.58s◆ (best)	50 tps◆ (best)	131K	$0.280 / $0.420
Friendlifriendli	Healthy	100%◆ (best)	99.88%	92.07%	92.07%	975ms	9.09s	30 tps	164K	$0.500 / $1.50
Googlegoogle-vertex	Healthy	100%◆ (best)	99.42%	99.70%	99.70%	1.32s	12.0s	19 tps	164K	$0.560 / $1.68
Novitanovita/fp8fp8	Healthy	100%◆ (best)	99.98%	99.99%◆ (best)	99.99%◆ (best)	1.35s	6.20s	26 tps	164K	$0.269 / $0.400
AtlasCloudatlas-cloud/fp8fp8	Healthy	99.95%	99.96%	99.96%	99.96%	1.58s	9.16s	22 tps	164K	$0.260 / $0.380
GMICloudgmicloud/fp8fp8	Healthy	99.94%	99.55%	99.57%	99.57%	2.04s	6.71s	5.0 tps	164K	$0.209 / $0.310◆ (best)
DigitalOceandigitalocean	Healthy	99.53%	96.46%	97.21%	97.21%	1.52s	14.4s	12 tps	164K	$0.250 / $0.800
Alibabaalibaba/fp8fp8	Healthy	98.91%	97.96%	98.40%	98.40%	844ms◆ (best)	12.0s	48 tps	131K	$0.370 / $1.11
StreamLakestreamlake/fp8fp8	Degraded	98.68%	97.02%	99.35%	99.35%	1.39s	9.96s	23 tps	128K	$0.214 / $0.322
Phalaphala	Healthy	98.32%	98.20%	99.06%	99.06%	2.64s	247s	5.0 tps	164K	$1.00 / $1.00
SiliconFlowsiliconflow/fp8fp8	Degraded	97.94%	94.89%	99.11%	99.11%	1.81s	11.7s	24 tps	164K	$0.259 / $0.420
Venicevenice	Down	16.33%	82.72%	75.07%	75.07%	3.90s	75.1s	3.0 tps	160K	$0.330 / $0.480
DeepInfradeepinfra/fp4fp4	No data	—	97.32%	93.77%	93.77%	999ms	9.90s	12 tps	164K	$0.260 / $0.380
SambaNovasambanova	No data	—	95.42%	96.95%	96.95%	—	—	—	33K	$3.00 / $4.50

Baidu100%◆ (best)
Healthy24h 99.91%
7d
99.94%
30d
99.94%
p50
999ms
p99
3.58s◆ (best)
Throughput
50 tps◆ (best)
Context
131K
$/Mtok
$0.280 / $0.420
baidu/fp8fp8
Friendli100%◆ (best)
Healthy24h 99.88%
7d
92.07%
30d
92.07%
p50
975ms
p99
9.09s
Throughput
30 tps
Context
164K
$/Mtok
$0.500 / $1.50
friendli
Google100%◆ (best)
Healthy24h 99.42%
7d
99.70%
30d
99.70%
p50
1.32s
p99
12.0s
Throughput
19 tps
Context
164K
$/Mtok
$0.560 / $1.68
google-vertex
Novita100%◆ (best)
Healthy24h 99.98%
7d
99.99%◆ (best)
30d
99.99%◆ (best)
p50
1.35s
p99
6.20s
Throughput
26 tps
Context
164K
$/Mtok
$0.269 / $0.400
novita/fp8fp8
AtlasCloud99.95%
Healthy24h 99.96%
7d
99.96%
30d
99.96%
p50
1.58s
p99
9.16s
Throughput
22 tps
Context
164K
$/Mtok
$0.260 / $0.380
atlas-cloud/fp8fp8
GMICloud99.94%
Healthy24h 99.55%
7d
99.57%
30d
99.57%
p50
2.04s
p99
6.71s
Throughput
5.0 tps
Context
164K
$/Mtok
$0.209 / $0.310◆ (best)
gmicloud/fp8fp8
DigitalOcean99.53%
Healthy24h 96.46%
7d
97.21%
30d
97.21%
p50
1.52s
p99
14.4s
Throughput
12 tps
Context
164K
$/Mtok
$0.250 / $0.800
digitalocean
Alibaba98.91%
Healthy24h 97.96%
7d
98.40%
30d
98.40%
p50
844ms◆ (best)
p99
12.0s
Throughput
48 tps
Context
131K
$/Mtok
$0.370 / $1.11
alibaba/fp8fp8
StreamLake98.68%
Degraded24h 97.02%
7d
99.35%
30d
99.35%
p50
1.39s
p99
9.96s
Throughput
23 tps
Context
128K
$/Mtok
$0.214 / $0.322
streamlake/fp8fp8
Phala98.32%
Healthy24h 98.20%
7d
99.06%
30d
99.06%
p50
2.64s
p99
247s
Throughput
5.0 tps
Context
164K
$/Mtok
$1.00 / $1.00
phala
SiliconFlow97.94%
Degraded24h 94.89%
7d
99.11%
30d
99.11%
p50
1.81s
p99
11.7s
Throughput
24 tps
Context
164K
$/Mtok
$0.259 / $0.420
siliconflow/fp8fp8
Venice16.33%
Down24h 82.72%
7d
75.07%
30d
75.07%
p50
3.90s
p99
75.1s
Throughput
3.0 tps
Context
160K
$/Mtok
$0.330 / $0.480
venice
DeepInfra—
No data24h 97.32%
7d
93.77%
30d
93.77%
p50
999ms
p99
9.90s
Throughput
12 tps
Context
164K
$/Mtok
$0.260 / $0.380
deepinfra/fp4fp4
SambaNova—
No data24h 95.42%
7d
96.95%
30d
96.95%
p50
—
p99
—
Throughput
—
Context
33K
$/Mtok
$3.00 / $4.50
sambanova

Measured 2m 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.