# Together AI says inference bottlenecks depend on traffic shape

> Original title: performance depends on the shape of your traffic

- Source: Together AI
- Published: 2026-11-12T04:01:21.000Z
- Bad Influence: https://badinfluence.news/items/j093etg5pm648m9gic7v80bbf
- Original: https://x.com/togethercompute/status/2120728631961927788

## Summary

Together AI says LLM performance depends on traffic shape, with short inputs and long outputs limited by decode speed and long prompts by prefill. It adds that tool-heavy agents are limited by end-to-end latency and repeated turns by token economics, so users should measure the bottleneck their workload creates.

## Post

performance depends on the shape of your traffic

short in / long out → decode speed. long prompts → prefill. tool-heavy agents → end-to-end latency. repeated turns → token economics

measure the bottleneck your workload actually creates

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHW5Xc8HWoAAuEuG.jpg&mode=thumb&exp=1794700800&sig=41e079dba68a55be)