1,000 output tokens

One trip to the other side = 1,000 output tokens · A round trip = 2,000

GPT-OSS 120BCerebras · 1,852 t/s

120Bparams ↗

GPT-OSS 20BGroq · 960 t/s

20Bparams ↗

Nemotron 3.5 LightningDeepInfra (pre-release) · 669 t/s

30B / 3Bparams ↗

GLM-5.2Baseten · 572 t/s

744Bparams ↗

Qwen3.8 Max PreviewAlibaba Cloud · ~465 t/s

~2.4Tparams ↗

Kimi K2.7 CodeCrusoe (NVFP4) · 462 t/s

1Tparams ↗

GLM-5.2Together AI · 439 t/s

744Bparams ↗

DeepSeek V4 FlashMakora · 232 t/s

284B / 13Bparams ↗

GPT-5.6 LunaOpenAI · 204 t/s

Undisclosedparams ↗

Gemini 3.5 FlashGoogle AI Studio · 195 t/s

Undisclosedparams ↗

GPT-5.6 TerraAmazon · 174 t/s

Undisclosedparams ↗

Kimi K3Fireworks · 162 t/s

2.8Tparams ↗

GPT-5.6 TerraOpenAI · 145 t/s

Undisclosedparams ↗

Muse Spark 1.2Meta · ~131 t/s

Undisclosedparams ↗

DeepSeek V4 FlashDeepSeek · 121 t/s

284B / 13Bparams ↗

Claude Sonnet 5Google · 87 t/s

Undisclosedparams ↗

Claude Fable 5Anthropic · 73 t/s

Undisclosedparams ↗

GPT-5.6 SolOpenAI · 71 t/s

Undisclosedparams ↗

Grok 4.5SpaceXAI · 60 t/s

Undisclosedparams ↗

Claude Opus 5Anthropic · 54 t/s

Undisclosedparams ↗