DeepSeek V3 0324: 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 | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Healthy | 100%◆ (best) | 99.93% | 99.94%◆ (best) | 99.94%◆ (best) | 396ms◆ (best) | 4.14s◆ (best) | 32 tps◆ (best) | 164K | $0.500 / $1.50 | |
| Healthy | 99.41% | 99.43% | 98.99% | 98.99% | 2.81s | 6.11s | 18 tps | 164K | $0.290 / $1.14 | |
| Healthy | 98.25% | 98.29% | 97.79% | 97.79% | 1.24s | 32.0s | 31 tps | 164K | $0.270 / $1.12 | |
| Healthy | 98.15% | 98.55% | 97.29% | 97.29% | 1.42s | 10.5s | 18 tps | 164K | $0.250 / $1.00 | |
| Degraded | 92.73% | 91.76% | 78.79% | 78.79% | 651ms | 23.9s | 22 tps | 164K | $0.240 / $0.900◆ (best) |
- Crusoe100%◆ (best)Healthy24h 99.93%
- 7d
- 99.94%◆ (best)
- 30d
- 99.94%◆ (best)
- p50
- 396ms◆ (best)
- p99
- 4.14s◆ (best)
- Throughput
- 32 tps◆ (best)
- Context
- 164K
- $/Mtok
- $0.500 / $1.50
crusoe/bf16bf16 - GMICloud99.41%Healthy24h 99.43%
- 7d
- 98.99%
- 30d
- 98.99%
- p50
- 2.81s
- p99
- 6.11s
- Throughput
- 18 tps
- Context
- 164K
- $/Mtok
- $0.290 / $1.14
gmicloud/fp8fp8 - Novita98.25%Healthy24h 98.29%
- 7d
- 97.79%
- 30d
- 97.79%
- p50
- 1.24s
- p99
- 32.0s
- Throughput
- 31 tps
- Context
- 164K
- $/Mtok
- $0.270 / $1.12
novita/fp8fp8 - SiliconFlow98.15%Healthy24h 98.55%
- 7d
- 97.29%
- 30d
- 97.29%
- p50
- 1.42s
- p99
- 10.5s
- Throughput
- 18 tps
- Context
- 164K
- $/Mtok
- $0.250 / $1.00
siliconflow/fp8fp8 - DeepInfra92.73%Degraded24h 91.76%
- 7d
- 78.79%
- 30d
- 78.79%
- p50
- 651ms
- p99
- 23.9s
- Throughput
- 22 tps
- Context
- 164K
- $/Mtok
- $0.240 / $0.900◆ (best)
deepinfra/fp4fp4
Measured 4m 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.
Throughput
Median tokens per second.