OpenAIAPI Host Benchmarks
Gpt 4o Mini — API Providers & Speed Matrix
Compare live throughput, TTFT latency distributions, prompt cache pricing, and hardware stacks across cloud hosting endpoints.
Fastest Provider
1125 t/s
Lowest Latency (TTFT)
0.09s P50
Best Price Value
$1.13 / $5
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
OpenAIOfficial | Cloud TPU / Nvidia H100 | 90 t/s | 0.8s (1.12s p90) | $2.5 | $10 | $0.25 | $4.38 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 378 t/s | 0.09s (0.12s p90) | $2.13 | $8.5 | $0.2 | $3.72 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 1125 t/s | 0.12s (0.16s p90) | $2.25 | $9 | — | $3.94 | 99.90% |
Together AI | Nvidia H100 SXM5 | 135 t/s | 0.64s (0.88s p90) | $1.63 | $7 | $0.15 | $2.97 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 117 t/s | 0.72s (1s p90) | $1.13 | $5 | $0.125 | $2.1 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 171 t/s | 0.56s (0.76s p90) | $1.5 | $6.5 | $0.15 | $2.75 | 99.99% |