Gemini 3.5 Flash-Lite

Adopt with guardrails

Why

GA pricing ($0.30/$2.50 per 1M tokens, 1M-token context) is roughly 80%+ cheaper than standard Gemini 3.5/3.6 Flash and independent analyst data confirms strong throughput and a competitive intelligence-for-price ratio, but the same analyst flags a high median time-to-first-token and Google's own materials frame Flash-Lite as the cost/speed tier below 3.6 Flash — guardrails (latency testing, scoping to batch/background paths) are warranted before treating it as a general-purpose default.

Next

Pilot Flash-Lite on one non-latency-critical, high-volume batch path (e.g. pv-tech-refresh corpus signal extraction or agentic document processing in apps/api) with side-by-side cost/latency/quality measurement against the current default model before any interactive-path adoption; re-test TTFT specifically inside workerd, since a 9.82s median first-token latency is risky for request-scoped Cloudflare Worker calls.