The Flash Family, Untangled
Google's "Flash" line is where most Gemini production traffic should live — frontier-adjacent quality at a fraction of Pro's cost. But there are several Flash tiers, and picking wrong wastes money. Here's the map (per 1M tokens, input / output):
| Model | Input | Output | Use it for |
|---|---|---|---|
| Gemini 3.5 Flash | ~$1.50 | ~$9 | Frontier-ish, coding/agents |
| Gemini 3 Flash | ~$0.50 | ~$3 | Strong value workhorse |
| Gemini 2.5 Flash | $0.30 | $2.50 | Balanced default |
| Gemini 3.1 Flash-Lite | ~$0.25 | ~$1.50 | Cheap, capable |
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 | Cheapest, high-volume |
Batch API halves any of these; context caching cuts cached input ~90%. On LinkModel they run behind one key with Claude, GPT, and DeepSeek, up to 30% below official.
Which Flash Should You Use?
- Gemini 3.5 Flash — the standout: it launched beating the pricier 3.1 Pro on several coding/agentic benchmarks at ~$1.50/$9. If you were reaching for Pro, try 3.5 Flash first.
- Gemini 2.5 Flash ($0.30/$2.50) — the reliable general-purpose default for chat, summarization, and extraction.
- Flash-Lite ($0.10/$0.40) — classification, routing, tagging, and high-volume simple transforms where a smaller model is fine.
The Cascade That Cuts Bills
The proven Gemini pattern: send everything to Flash-Lite or Flash first, escalate only the small fraction that needs deeper reasoning to Pro. Google's own guidance is that this cuts spend substantially with little quality loss. Two cautions: the 2.5 family bills thinking tokens in output (keep reasoning off by default), and Pro tiers moved to paid-only in April 2026. More in how to reduce AI API costs.
How to Call It
curl -X POST https://api.linkmodel.ai/api/v1/... \
-H "Authorization: Bearer $LINKMODEL_API_KEY" -H "Content-Type: application/json" \
-d '{ "model": "gemini-3.5-flash", "messages": [{"role":"user","content":"Summarize and tag these support tickets: ..."}] }'Confirm the exact chat endpoint and schema in the docs.
Flash vs the Field
Gemini 3.5 Flash (~$1.50/$9) is often the best value at quality across all vendors — cheaper than Claude Sonnet ($3/$15) and GPT-5.4 ($2.50/$15), while Flash-Lite rivals DeepSeek V4 Flash on input. See cheapest LLM API and best LLM API.
Bottom Line
Use Gemini 3.5 Flash when you'd otherwise pay for Pro, 2.5 Flash as the everyday default, and Flash-Lite for cheap high-volume work — then cascade to capture the savings.
Start free with a $1 credit and benchmark the Flash tiers on your traffic.