Google dropped Gemini 3.6 Flash this week.
By John Aspinall ✱ · July 24, 2026 · Curated by George's Blog
Google dropped Gemini 3.6 Flash this week.
Ignore all the model-launch theater.
The useful part is the economics.
3.6 Flash is the new workhorse.
Google’s own numbers:
17% fewer output tokens than 3.5 Flash (Artificial Analysis)
Fewer tool calls on multi-step jobs
Output price cut to $7.50 per million tokens (from $9 on 3.5 Flash).
Input stays $1.50/M.
Same class of work.
Cheaper per finished task.
That’s the only metric that matters if you’re running operators, not demos.
Where I’d actually point it:
❶ Volume lanes - batch title rewrites, Q&A banks, listing audits, competitor scrapes, weekly report assembly. High call count. Low regret if one answer is mid.
❷ Agent steps - the multi-hop stuff in your stack (research → draft → format → ship). Google is optimizing for fewer reasoning loops and tool calls. That’s your overnight jobs and cron agents.
❸ Multimodal grunt work - chart reads, doc dumps, screenshot → structured notes. Good enough for first pass. Not final creative judgment.
❹ Flash-Lite for the firehose - they also shipped 3.5 Flash-Lite at $0.30 / $2.50 per million and ~350 tok/s. Classification, extraction, search-y subagents. Don’t put your strategy brain on this model.
What I would NOT put on 3.6 Flash without a human gate:
Final hero strategy.
Client-facing claims.
Anything where a wrong number costs money.
⤷ Pricing reviews. Legal-adjacent copy.
Some builders on X already say vision/object detection got lazier than 3.5. Fine for operator text pipelines. Don’t trust it blindly on “find every defect in this image stack.”
Router rule I use:
Cheap + fast for volume.
Expensive + slow for judgment.
3.6 Flash is a volume chip. Route it there.
Keep the hard calls on a stronger model (or a human).
If your bill is still 100% frontier model for every listing rewrite and every status email, you’re not buying intelligence. You’re buying habit.
Swap the volume lane this week.
Measure cost per finished artifact, not vibes on a benchmark chart.
3.5 Pro is still in partner testing.
Gemini 4 is already in pre-training.
Don’t wait for either to fix your routing table.