GoogleAPI Host BenchmarksEstimated
Gemma 3 12b — API Providers & Speed Matrix
Modelled throughput, TTFT latency and prompt-cache pricing estimates across cloud hosting endpoints, derived from official list prices and public benchmarks.
Fastest provider
1500 t/s
Lowest latency (TTFT)
90 ms P50
Best price
$0.56 / $2.5
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
GoogleOfficial | Cloud TPU / Nvidia H100 | 120 t/s | 800 ms (1.12 s p90) | $1.25 | $5 | $0.125 | $2.19 | 99.99% |
| Groq LPU v2 (SRAM) | 504 t/s | 90 ms (120 ms p90) | $1.06 | $4.25 | $0.1 | $1.86 | 99.95% | |
| CS-3 Wafer-Scale Engine | 1500 t/s | 120 ms (160 ms p90) | $1.13 | $4.5 | — | $1.97 | 99.90% | |
| Nvidia H100 SXM5 | 180 t/s | 640 ms (880 ms p90) | $0.81 | $3.5 | $0.075 | $1.48 | 99.95% | |
| Nvidia H100 / L40S | 156 t/s | 720 ms (1.00 s p90) | $0.56 | $2.5 | $0.063 | $1.04 | 99.90% | |
| Nvidia B200 / H100 | 228 t/s | 560 ms (760 ms p90) | $0.75 | $3.25 | $0.075 | $1.38 | 99.99% |
Figures are modelled estimates based on published list prices and public throughput benchmarks — not measured on live endpoints. Verify current pricing and terms with each host before committing.