DeepSeekAPI Host Benchmarks
DeepSeek V4 Flash Vision — API Providers & Speed Matrix
Compare live throughput, TTFT latency distributions, prompt cache pricing, and hardware stacks across cloud hosting endpoints.
Fastest Provider
1500 t/s
Lowest Latency (TTFT)
0.09s P50
Best Price Value
$0.2 / $0.66
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
DeepSeekOfficial | Cloud TPU / Nvidia H100 | 120 t/s | 1.08s (1.51s p90) | $0.44 | $1.32 | $0.044 | $0.66 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 504 t/s | 0.09s (0.12s p90) | $0.37 | $1.12 | $0.035 | $0.56 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 1500 t/s | 0.12s (0.16s p90) | $0.4 | $1.19 | — | $0.6 | 99.90% |
Together AI | Nvidia H100 SXM5 | 180 t/s | 0.86s (1.19s p90) | $0.29 | $0.92 | $0.026 | $0.45 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 156 t/s | 0.97s (1.35s p90) | $0.2 | $0.66 | $0.022 | $0.32 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 228 t/s | 0.76s (1.03s p90) | $0.26 | $0.86 | $0.026 | $0.41 | 99.99% |