Tokens per second

What it means

Higher values mean faster responses. In operations workflows, tokens per second often shapes performance and reliability.

How it works

Operations covers latency, throughput, and cost. Systems often use caching, batching, and monitoring to scale reliably.

Why it matters

Operational choices impact cost, latency, and reliability.

Common use cases

• Reducing time-to-first-token with streaming.
• Managing costs with token budgets and caching.
• Tracking usage and errors with logs and metrics.

Example

A model that outputs 120 tokens per second.

Pitfalls and tips

Ignoring limits can cause timeouts or rate limiting. Set budgets and monitor usage to avoid surprises.

In BoltAI

In BoltAI, this shows up in performance, logging, or usage views.