Token speed fails to predict real task time or cost, a 13-model probe finds
According-Floor5177 · reddit · 2026-07-23
A Reddit user argues that tokens per second is a poor proxy for how fast or cheaply a model finishes a real task.
- In one probe across 13 models, KAT-Coder was third in token output speed but finished tenth because it spent 5,536 tokens, while GPT-5.5 used only 1,777.
- The same probe found provider variance: Kimi K2.7 ranged from 223 tok/s on Together to 28 tok/s on DeepInfra, and GLM-5.2 landed on six different providers in six runs.
- The takeaway is that wall-clock time and end-to-end cost matter more than raw generation speed, and provider routing can materially change results.
More from Infra
- Ubicloud publishes "Fixing Tool-Call Accuracy in vLLM" — petrusenko_max · 2026-10-26
- Meta approves nonvoting Class C shares as 2026 capex guidance hits $135B–$145B — thione · 2026-10-26
- Unsloth Ships MiniMax-M3.1-Flash Quants: 2-bit Retains 91.5% Accuracy, Runs in 150GB RAM — danielhanchen · 2026-10-26
- Engineer with thousands of fabricated devices pours cold water on photonic crystal modulator yields — jwt0625 · 2026-10-26
- Anthropic S-1 reveals 48 cents of compute cost per dollar of Q2 revenue — nmasc_ · 2026-10-26
- Persistent AI joins UR network to route private AI inference with encrypted, IP-hidden traffic — 0xSammy · 2026-10-26