ModelRefs / First-Token Latency Methodology and Evidence
First-Token Latency Methodology and Evidence
Time from request to first streamed token (lower is better). Methodology, dataset provenance, and citation sources for First-Token Latency.
Overview
Time from request to first streamed token (lower is better).
How it is measured: Median over 100 requests, 256-token prompt, US-East endpoint.
How this benchmark is scored
| Category | latency |
|---|---|
| Maximum score | 2000 ms |
| Direction | Lower is better |
Primary source: https://artificialanalysis.ai/
Published results
| Model | Score | Run date | Source |
|---|---|---|---|
| Llama 4 Scout | 290 | 2026-05-01 | Aggregated public reports |
| GPT-5 Mini | 320 | 2026-05-01 | Aggregated public reports |
| Mistral Large 2 | 410 | 2026-05-01 | Aggregated public reports |
| Gemini 3 Pro | 480 | 2026-05-01 | Aggregated public reports |
| GPT-5 | 540 | 2026-05-01 | Aggregated public reports |
| Claude Sonnet 4.5 | 610 | 2026-05-01 | Aggregated public reports |
Each score reflects the protocol and date of its own source run. Results from different harnesses are not directly comparable.
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to First-Token Latency Methodology and Evidence.
Frequently asked questions
What does the First-Token Latency benchmark measure?
Time from request to first streamed token (lower is better).
What is the methodology?
Median over 100 requests, 256-token prompt, US-East endpoint.