MicrosoftAPI Host Benchmarks
Phi-4 Multimodal — API Providers & Speed Matrix
Compare live throughput, TTFT latency distributions, prompt cache pricing, and hardware stacks across cloud hosting endpoints.
Fastest Provider
1063 t/s
Lowest Latency (TTFT)
0.09s P50
Best Price Value
$0.04 / $0.16
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
MicrosoftOfficial | Cloud TPU / Nvidia H100 | 85 t/s | 0.8s (1.12s p90) | $0.08 | $0.32 | $0.008 | $0.14 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 357 t/s | 0.09s (0.12s p90) | $0.07 | $0.27 | $0.006 | $0.12 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 1063 t/s | 0.12s (0.16s p90) | $0.07 | $0.29 | — | $0.13 | 99.90% |
Together AI | Nvidia H100 SXM5 | 128 t/s | 0.64s (0.88s p90) | $0.05 | $0.22 | $0.005 | $0.09 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 111 t/s | 0.72s (1s p90) | $0.04 | $0.16 | $0.004 | $0.07 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 162 t/s | 0.56s (0.76s p90) | $0.05 | $0.21 | $0.005 | $0.09 | 99.99% |