OpenAIAPI Host Benchmarks
GPT-5.6 Luna — API Providers & Speed Matrix
Compare live throughput, TTFT latency distributions, prompt cache pricing, and hardware stacks across cloud hosting endpoints.
Fastest Provider
2188 t/s
Lowest Latency (TTFT)
0.09s P50
Best Price Value
$0.09 / $0.6
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
OpenAIOfficial | Cloud TPU / Nvidia H100 | 175 t/s | 135.99s (190.39s p90) | $0.2 | $1.2 | $0.02 | $0.45 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 735 t/s | 0.09s (0.12s p90) | $0.17 | $1.02 | $0.016 | $0.38 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 2188 t/s | 0.12s (0.16s p90) | $0.18 | $1.08 | — | $0.41 | 99.90% |
Together AI | Nvidia H100 SXM5 | 263 t/s | 108.79s (149.59s p90) | $0.13 | $0.84 | $0.012 | $0.31 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 228 t/s | 122.39s (169.99s p90) | $0.09 | $0.6 | $0.01 | $0.22 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 333 t/s | 95.19s (129.19s p90) | $0.12 | $0.78 | $0.012 | $0.29 | 99.99% |