Gemma 4 31B: 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
Cerebrascerebras/fp16fp16	Healthy	100%◆ (best)	100%	100%◆ (best)	100%◆ (best)	425ms	10.6s	163 tps	131K	$0.990 / $1.49
DeepInfradeepinfra/fp8fp8	Healthy	100%◆ (best)	98.61%	91.14%	91.14%	431ms	5.29s	40 tps	262K	$0.130 / $0.380
ModelRunmodelrun/fp4fp4	Healthy	100%◆ (best)	99.92%	99.67%	99.67%	87ms◆ (best)	1.58s◆ (best)	164 tps◆ (best)	262K	$0.750 / $1.00
Novitanovita/bf16bf16	Healthy	100%◆ (best)	99.11%	80.32%	80.32%	974ms	8.63s	9.0 tps	262K	$0.140 / $0.400
OpenInferenceopen-inference/bf16bf16	Healthy	100%◆ (best)	98.10%	97.61%	97.61%	973ms	30.6s	18 tps	262K	$0.080 / $0.350◆ (best)
SiliconFlowsiliconflow/fp8fp8	Degraded	100%◆ (best)	88.66%	94.18%	94.18%	3.88s	51.2s	16 tps	262K	$0.130 / $0.400
Venicevenice/bf16bf16	Healthy	100%◆ (best)	99.76%	99.74%	99.74%	1.13s	18.1s	34 tps	256K	$0.120 / $0.360
DeepInfradeepinfra/turbofp4	Healthy	99.93%	99.84%	99.54%	99.54%	900ms	7.90s	20 tps	262K	$0.090 / $0.340
Friendlifriendli	Healthy	99.92%	99.90%	99.58%	99.58%	626ms	7.83s	65 tps	262K	$0.140 / $0.400
Crusoecrusoe	Healthy	99.60%	98.74%	98.65%	98.65%	758ms	42.6s	23 tps	262K	$0.140 / $0.400
CoreWeavecoreweave/bf16bf16	Healthy	97.88%	98.27%	98.49%	98.49%	1.02s	26.6s	28 tps	262K	$0.100 / $0.340
Parasailparasail/fp8fp8	Healthy	95.90%	99.56%	99.28%	99.28%	564ms	10.8s	23 tps	262K	$0.150 / $0.400
DeepInfradeepinfra/ultrafp8	Down	84.85%	77.66%	88.77%	88.77%	2.98s	16.3s	22 tps	131K	$0.270 / $0.760
Chuteschutes/fp4fp4	Down	—	90.47%	64.48%	64.48%	3.05s	117s	10 tps	131K	$0.120 / $0.370
Phalaphala	Down	—	73.32%	84.66%	84.66%	1.66s	12.8s	30 tps	262K	$0.150 / $0.460
SambaNovasambanova	Degraded	—	98.15%	96.94%	96.94%	2.97s	27.1s	44 tps	262K	$0.380 / $1.15
Togethertogether	No data	—	92.88%	95.44%	95.44%	259ms	22.2s	50 tps	262K	$0.280 / $0.860

Cerebras100%◆ (best)
Healthy24h 100%
7d
100%◆ (best)
30d
100%◆ (best)
p50
425ms
p99
10.6s
Throughput
163 tps
Context
131K
$/Mtok
$0.990 / $1.49
cerebras/fp16fp16
DeepInfra100%◆ (best)
Healthy24h 98.61%
7d
91.14%
30d
91.14%
p50
431ms
p99
5.29s
Throughput
40 tps
Context
262K
$/Mtok
$0.130 / $0.380
deepinfra/fp8fp8
ModelRun100%◆ (best)
Healthy24h 99.92%
7d
99.67%
30d
99.67%
p50
87ms◆ (best)
p99
1.58s◆ (best)
Throughput
164 tps◆ (best)
Context
262K
$/Mtok
$0.750 / $1.00
modelrun/fp4fp4
Novita100%◆ (best)
Healthy24h 99.11%
7d
80.32%
30d
80.32%
p50
974ms
p99
8.63s
Throughput
9.0 tps
Context
262K
$/Mtok
$0.140 / $0.400
novita/bf16bf16
OpenInference100%◆ (best)
Healthy24h 98.10%
7d
97.61%
30d
97.61%
p50
973ms
p99
30.6s
Throughput
18 tps
Context
262K
$/Mtok
$0.080 / $0.350◆ (best)
open-inference/bf16bf16
SiliconFlow100%◆ (best)
Degraded24h 88.66%
7d
94.18%
30d
94.18%
p50
3.88s
p99
51.2s
Throughput
16 tps
Context
262K
$/Mtok
$0.130 / $0.400
siliconflow/fp8fp8
Venice100%◆ (best)
Healthy24h 99.76%
7d
99.74%
30d
99.74%
p50
1.13s
p99
18.1s
Throughput
34 tps
Context
256K
$/Mtok
$0.120 / $0.360
venice/bf16bf16
DeepInfra99.93%
Healthy24h 99.84%
7d
99.54%
30d
99.54%
p50
900ms
p99
7.90s
Throughput
20 tps
Context
262K
$/Mtok
$0.090 / $0.340
deepinfra/turbofp4
Friendli99.92%
Healthy24h 99.90%
7d
99.58%
30d
99.58%
p50
626ms
p99
7.83s
Throughput
65 tps
Context
262K
$/Mtok
$0.140 / $0.400
friendli
Crusoe99.60%
Healthy24h 98.74%
7d
98.65%
30d
98.65%
p50
758ms
p99
42.6s
Throughput
23 tps
Context
262K
$/Mtok
$0.140 / $0.400
crusoe
CoreWeave97.88%
Healthy24h 98.27%
7d
98.49%
30d
98.49%
p50
1.02s
p99
26.6s
Throughput
28 tps
Context
262K
$/Mtok
$0.100 / $0.340
coreweave/bf16bf16
Parasail95.90%
Healthy24h 99.56%
7d
99.28%
30d
99.28%
p50
564ms
p99
10.8s
Throughput
23 tps
Context
262K
$/Mtok
$0.150 / $0.400
parasail/fp8fp8
DeepInfra84.85%
Down24h 77.66%
7d
88.77%
30d
88.77%
p50
2.98s
p99
16.3s
Throughput
22 tps
Context
131K
$/Mtok
$0.270 / $0.760
deepinfra/ultrafp8
Chutes—
Down24h 90.47%
7d
64.48%
30d
64.48%
p50
3.05s
p99
117s
Throughput
10 tps
Context
131K
$/Mtok
$0.120 / $0.370
chutes/fp4fp4
Phala—
Down24h 73.32%
7d
84.66%
30d
84.66%
p50
1.66s
p99
12.8s
Throughput
30 tps
Context
262K
$/Mtok
$0.150 / $0.460
phala
SambaNova—
Degraded24h 98.15%
7d
96.94%
30d
96.94%
p50
2.97s
p99
27.1s
Throughput
44 tps
Context
262K
$/Mtok
$0.380 / $1.15
sambanova
Together—
No data24h 92.88%
7d
95.44%
30d
95.44%
p50
259ms
p99
22.2s
Throughput
50 tps
Context
262K
$/Mtok
$0.280 / $0.860
together

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.