OpenAIAPI Host Benchmarks
GPT-5.3 Codex — API Providers & Speed Matrix
Compare live throughput, TTFT latency distributions, prompt cache pricing, and hardware stacks across cloud hosting endpoints.
Fastest Provider
1475 t/s
Lowest Latency (TTFT)
0.09s P50
Best Price Value
$0.79 / $7
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
OpenAIOfficial | Cloud TPU / Nvidia H100 | 118 t/s | 0.8s (1.12s p90) | $1.75 | $14 | $0.175 | $4.81 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 496 t/s | 0.09s (0.12s p90) | $1.49 | $11.9 | $0.14 | $4.09 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 1475 t/s | 0.12s (0.16s p90) | $1.57 | $12.6 | — | $4.33 | 99.90% |
Together AI | Nvidia H100 SXM5 | 177 t/s | 0.64s (0.88s p90) | $1.14 | $9.8 | $0.105 | $3.31 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 153 t/s | 0.72s (1s p90) | $0.79 | $7 | $0.088 | $2.34 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 224 t/s | 0.56s (0.76s p90) | $1.05 | $9.1 | $0.105 | $3.06 | 99.99% |