BENCHMARKFirst benchmarks recorded: MiniMax M3 → Cortecs → ?4hBENCHMARKFirst benchmarks recorded: Kimi K3 → Cortecs → ?4hBENCHMARKFirst benchmarks recorded: DeepSeek V4 Pro → Cortecs → ?4hBENCHMARKFirst benchmarks recorded: GLM-4.7 → Cortecs → ?4hBENCHMARKFirst benchmarks recorded: GLM-5.2 → Cortecs → ?4h

DeepSeek-V3

6 providers · 671B MoE · 128k context · 3 regions · all models →

DeepSeek-V3 across 6 providers

Same model. Different infrastructure. DEV DATA
TTFT spread2.0×Throughput spread2.1×Output price2.8×Tool reliability4.1 pts
TTFT340 ms693 msbest: Nebius
Throughput72 tok/s153 tok/sbest: Fireworks
Output price$0.39/M$1.10/Mbest: SiliconFlow
Reliability96.6%99.0%best: Fireworks
Tool success94.8%98.9%best: Fireworks

DeepSeek-V3 — tool-calling benchmark

6 providers · 94.8% → 98.9% · 4.1 pt spread · DEV DATA
#ProviderToolsΔ bestRel.JSONTTFTP95Tok/s$ Out
01Fireworks98.9%—99.0%99.7%368 ms1272 ms153$1.10
02OpenRouter98.6%−0.397.6%98.7%453 ms1465 ms131$1.10
03Nebius97.7%−1.298.9%99.1%340 ms1257 ms116$0.50
04Together96.8%−2.197.8%98.3%473 ms1600 ms107$0.90
05DeepInfra96.1%−2.897.4%98.0%604 ms1696 ms72$0.45
06SiliconFlow94.8%−4.196.6%97.7%693 ms2048 ms75$0.39