Gemini 3.8 Flash is cheap. Until 1 January
Google's Gemini 3.8 Flash is fast, capable and cheap, which is why small models are the ones most people actually use. But its introductory price doubles on 1 January, and the Cyber version is invite-only.
On 2 September Google released Gemini 3.8 Flash, the latest version of its fast, cheap AI model, alongside a locked-down sibling called Gemini 3.8 Flash Cyber that hunts for security holes in software. It's the third Flash release in roughly three months. It also comes with a price tag that has an expiry date.
Developers pay $0.75 per million tokens of input and $3.75 per million tokens of output (a token is roughly three-quarters of a word). That's an introductory rate. From 1 January 2027 it doubles, to $1.50 and $7.50.
The small model is the one you actually use
Flagship models such as Gemini Pro get the headlines. Flash models do the work. They sit behind chatbots, search summaries and apps that call an AI model many times a minute, where speed and cost matter most.
Google's own UK subscription page makes the point. Free Gemini users in Britain get "3.6 Flash", with only "varying access" to the bigger Pro model. For most people, the Flash tier is Gemini.
Google calls 3.8 Flash its "most intelligent workhorse model", with gains over 3.7 Flash in software engineering, agent tasks and multi-step reasoning. It reports 54.9% on HLE-Verified, a tough reasoning test, and says 3.8 Flash beats most larger frontier models on the DeepSWE coding benchmark. VentureBeat notes it ranked seventh in the crowd-voted Text Arena leaderboard at launch. Those are mostly Google's own figures.
Cheaper per word, hungrier per job
There's a catch in the small print. Google says 3.8 Flash "works harder" on complex tasks, taking extra reasoning steps and calling tools repeatedly. As 9to5Google reports, Google admits this can mean using more tokens to get a better answer.
So the per-token price is only half the story. A model that is cheap per word but chattier per job can cost more than you planned. Anyone moving a busy app from 3.7 to 3.8 Flash should measure the real bill, not the rate card.
Introductory prices, and then the doubling
Then there's the clock. Price tracker TokenCost points out that the 31 December end date isn't tied to 3.8 Flash at all. It's a calendar deadline shared by 3.6, 3.7 and 3.8 Flash, so each new release gets a shorter discount: 3.7 Flash had 141 days of it, and 3.8 Flash gets 121. On 1 January, every tier doubles at once.
Launch cheap, build a habit, then raise the price is a familiar pattern. We saw a version of it when DeepSeek put up its prices. Google has at least been upfront about the date. But a business that plans its 2027 budget on today's numbers is in for a surprise.
The Cyber twin you can't have
Gemini 3.8 Flash Cyber is a version tuned to find and fix software vulnerabilities. Google says it has "more permissive safeguards", which is exactly why you can't just sign up. Access goes through Google's Fairwind Program, aimed at "trusted government authorities, as well as critical infrastructure operators and software maintainers".
The claims are bold. Google says Chrome's security team found that 3.8 Flash Cyber "produced 2.6 times more correct patches to vulnerabilities in Chrome than the best commercial models that are much larger". VentureBeat reports an 86.2% score on the CyberGym benchmark. Neither has been independently replicated yet.
What it means in the UK
For UK consumers, 3.8 Flash sits behind a paywall. In the Gemini app it's available to Google AI Pro and Ultra subscribers. Google's UK prices are £18.99 a month for AI Pro and from £79.99 for Ultra, with a cheaper £4.49 AI Plus tier and the free plan on the older 3.6 Flash. Robby Stein, Google's VP of Product for Search, said it was "available today in AI Mode for Google AI Pro & Ultra subs around the world", so UK subscribers got it on day one.
For UK developers and businesses, the API is billed in US dollars. QuantSpark's UK pricing guide puts the introductory rate at roughly 56p per million input tokens and £2.82 per million output at current exchange rates, and flags the January doubling.
To be fair to Google
This is a genuinely capable model at a low price, and Google has published the end date for the discount rather than burying it. Gating the Cyber version is responsible: a tool that is good at finding holes is also good at exploiting them.
Mind the hype. The Flash is cheap until New Year's Day. Set a reminder.
Sources: Google, VentureBeat, 9to5Google, Thurrott, Search Engine Journal, TokenCost, Google AI plans (UK), QuantSpark