v148 · AI · JavaScript

Quality Control Panel

Benchmark the Prompt API across four dimensions: first-token latency, total response time, context window usage, and output quality. Tune temperature and topK, then see how those parameters affect speed and coherence.

— Avg first-token (ms)

— Avg total (ms)

— Context tokens used

— Inference mode

Parameters

Temperature

0.50

Top-K

Benchmark suite

// Measuring Prompt API performance
const session = await window.ai.languageModel.create({
  temperature: 0.5,
  topK: 40,
});

// First-token latency: time to first streamed chunk
const firstToken = { time: null };
const t0 = performance.now();
const stream = await session.promptStreaming(prompt);
let result = '';

for await (const chunk of stream) {
  if (!firstToken.time) firstToken.time = performance.now() - t0;
  result = chunk;
}
const totalMs = performance.now() - t0;

// Context window: session.tokensSoFar / session.maxTokens
console.log({ firstToken, totalMs, tokensUsed: session.tokensSoFar });

implementation reference

Need the exact API surface, compatibility boundaries, errors, lifecycle, and source links? Read the matching gendn reference ↗

see also

implementation reference