Cook with AI Enrollment closed
AI & LLMs Runs locally

LLM Latency Calculator

Estimate LLM response time by output length — compare time to first token and total wait across models.

ELI5

It estimates how long users will wait for an AI answer.

A little more detail

What this tool does

AI responses have two useful timing numbers. Time to first token measures when text starts appearing. Generation speed determines when the full answer finishes. This tool estimates both.

Use it to
  • Compare how fast several models feel to a user
  • Estimate total wait time for short and long answers
  • Set sensible output limits for interactive features
Interactive workspace Results update as you type
Output length

Tokens in the model's reply (not your prompt).

Highlight model
Estimated response time

Sorted fastest to slowest. TTFT = time to first token; Gen = remaining generation time.

Model Provider TTFT Gen Total Relative
Streaming vs waiting

Everything runs in your browser. Numbers are rough mid-2026 estimates — not benchmarks from your exact prompt or region.

Latency is the UX cost nobody puts in a spreadsheet. This calculator combines time-to-first-token (TTFT) with output throughput to estimate total wait time: total ≈ TTFT + outputTokens ÷ tokens/sec.

Figures are ballpark mid-2026 estimates from public benchmarks and vendor docs. They vary with prompt size, queue depth, caching, region, and whether the model is warm. Use them to compare orders of magnitude — not to promise SLAs.

Groq and similar speed-focused hosts trade model size and reasoning depth for raw tokens/sec. Frontier models on OpenAI or Anthropic are slower per token but often need fewer tokens to finish the job — latency and quality are not the same tradeoff.