AlibabaAPI Host Benchmarks
Qwen3 Vl 235b A22b Instruct — 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
$0.23 / $1
Blended Pricing Workload Calculator
Switch between standard conversational ratios, agentic caching loops, and heavy RAG workloads.
| Provider & Host | Hardware Stack | Output Speed (TPS) | Latency TTFT (P50) | Input Price ($/1M) | Output Price ($/1M) | Cache Read ($/1M) | Blended Price ($/1M) | SLA |
|---|---|---|---|---|---|---|---|---|
AlibabaOfficial | Cloud TPU / Nvidia H100 | 85 t/s | 0.8s (1.12s p90) | $0.5 | $2 | $0.05 | $0.88 | 99.99% |
Groq Cloud | Groq LPU v2 (SRAM) | 357 t/s | 0.09s (0.12s p90) | $0.42 | $1.7 | $0.04 | $0.74 | 99.95% |
Cerebras Cloud | CS-3 Wafer-Scale Engine | 1063 t/s | 0.12s (0.16s p90) | $0.45 | $1.8 | — | $0.79 | 99.90% |
Together AI | Nvidia H100 SXM5 | 128 t/s | 0.64s (0.88s p90) | $0.33 | $1.4 | $0.03 | $0.6 | 99.95% |
DeepInfra | Nvidia H100 / L40S | 111 t/s | 0.72s (1s p90) | $0.23 | $1 | $0.025 | $0.42 | 99.90% |
Fireworks AI | Nvidia B200 / H100 | 162 t/s | 0.56s (0.76s p90) | $0.3 | $1.3 | $0.03 | $0.55 | 99.99% |