Google's Gemini 3.8 Flash is cheap until New Year's Eve

The price is the detail to put in your diary. Google lists 3.8 Flash at $0.75 per million input tokens and $3.75 per million output tokens as an introductory rate that runs until 31 December 2026. From 1 January 2027 it becomes $1.50 and $7.50.

3.8 Flash Cyber is only available to vetted defenders, including government authorities, critical infrastructure operators and software maintainers, through Google's Fairwind Program. On an internal Google benchmark covering 20 programming languages, it reached a success rate above 70%.

Why the price change matters

Flash models tend to end up in the busy, unglamorous parts of a product: sorting support tickets, pulling details out of documents, tagging leads, first-pass agents. They run thousands of times a day, so doubling the token price flows straight into your gross margin.

Here's what that looks like on a real job. Say you classify 50,000 inbound emails a month, each using about 2,000 input tokens and 500 output tokens. That's 100 million input tokens and 25 million output tokens a month (FER calculations from Google's list prices):

• Until 31 December: $75 plus about $94, so roughly $169 a month.
• From 1 January: $150 plus $187.50, so roughly $338 a month.

Not a crisis at that size. But multiply it by a few features, or by every client you've built one for, and it's a line your finance person will ask about in February.

There's a second catch. The Register reports that 3.8 Flash costs about 40% more per task than its similarly priced predecessor, because it tends to write more tokens and take more turns when it's acting as an agent. So the price per million tokens isn't the price per job. Measure what a finished task costs you, not what the rate card says.

What to do before January

1. Find every feature, automation or client build that calls Gemini Flash. Check the model name in the config, not just the docs.
2. Cost each one at the 2027 price using last month's real volume, and the real tokens per task from your logs rather than an estimate.
3. For the biggest one, test whether a smaller or cheaper model does the job as well on 50 real examples.
4. If you charge clients for AI usage, check whether your contract lets you pass the increase on, and tell them before January rather than after.
5. Put a reminder in your diary for mid-December to check whether Google has changed the plan.

If the numbers still work at double the price, keep it and move on. The point is to decide in October, not to discover it on your January invoice.