DeepSeekAPI Host Benchmarks
Deepseek V2.5 1210 — API Providers & Speed Matrix
Compare live throughput, TTFT latency distributions, prompt cache pricing, and hardware stacks across cloud hosting endpoints.
Fastest Provider
625 t/s
Lowest Latency (TTFT)
0.09s P50
Best Price Value
$0.12 / $0.55
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
DeepSeekOfficial | Cloud TPU / Nvidia H100 | 50 t/s | 0.8s (1.12s p90) | $0.27 | $1.1 | $0.027 | $0.48 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 210 t/s | 0.09s (0.12s p90) | $0.23 | $0.94 | $0.022 | $0.41 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 625 t/s | 0.12s (0.16s p90) | $0.24 | $0.99 | — | $0.43 | 99.90% |
Together AI | Nvidia H100 SXM5 | 75 t/s | 0.64s (0.88s p90) | $0.18 | $0.77 | $0.016 | $0.33 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 65 t/s | 0.72s (1s p90) | $0.12 | $0.55 | $0.014 | $0.23 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 95 t/s | 0.56s (0.76s p90) | $0.16 | $0.72 | $0.016 | $0.3 | 99.99% |