Three Weeks, Not Three Months
On September 2, 2026, Google DeepMind published Gemini 3.8 Flash on its official blog, twenty days after Gemini 3.7 Flash had shipped on August 13. A few weeks ago this would have been described as a routine update. At three weeks apart, it describes something else: not an update cycle, but a new competitive cadence.
9to5Google picked up the release the same day, noting that DeepMind also added new agentic video-understanding capabilities to Gemini and introduced a security product called Fairwind. Three announcements bundled into one release is unusual for a company that otherwise keeps a steadier pace. That pattern is hard to miss for any team already comparing cloud AI vendors on committed annual spend.
Fairwind Arrives the Same Week as a Warning Signal
Alongside the Flash refresh, DeepMind introduced Fairwind, described as a proactive cyber-defense program for governments and enterprises. Few technical details are public yet: DeepMind has not explained how Fairwind works, what it monitors, or how enterprises would deploy it. What is public is the timing. That gap between what DeepMind is willing to announce and what it is willing to detail stands out, given how publicly the company has marketed nearly every other part of this release.
The same week, Servola reported separately that OpenAI's unreleased Astra model had crossed into what OpenAI's own classification calls critical offensive cyber capability. DeepMind naming a defensive program in the very week a rival model crosses an offensive threshold is not something an attentive reader should let pass unremarked. Whether Fairwind is a direct response or the overlap is coincidence, the practical effect is the same: AI vendors are starting to sell defense against AI-enabled threats as a named, marketed product.
The Pricing Lock-In Problem
Gemini 3.7 Flash launched on August 13 at $0.75 per million input tokens and $3.75 per million output tokens, roughly 0.70 and 3.45 euros at an approximate conversion, an introductory rate DeepMind confirmed was exactly half of the $1.50 and $7.50 rate scheduled to take effect January 1, 2027. Gemini 3.8 Flash now ships at its own introductory pricing, confirmed to hold through December 31, 2026, though DeepMind has not published the exact per-token figures.
| Model | Launch date | Pricing | Note |
|---|---|---|---|
| Gemini 3.7 Flash | 13 August 2026 | $0.75 / $3.75 per million tokens (input/output) | Introductory rate, half of the $1.50 / $7.50 rate due 1 January 2027 |
| Gemini 3.8 Flash | 2 September 2026 | Introductory pricing (exact rate not disclosed) | Holds through 31 December 2026 |
| Rate cliff | 1 January 2027 | $1.50 / $7.50 per million tokens | Previously announced end date for the 3.7 Flash introductory rate |
For an EU or UK business budgeting a production workload against Gemini pricing, the arithmetic used to be simple: lock in the introductory rate, plan around the known cliff on January 1, 2027. A three-week refresh cycle breaks that simplicity. Any rate negotiated today can be undercut, replaced, or restructured by DeepMind's own next release before a typical annual contract term is even half finished. Businesses that treated the January cliff as the only date to plan around now have a second, moving target to track.
Defense-Against-AI Becomes a Product Category
Fairwind is not the first defensive AI security tool on the market, but a major lab naming and marketing one alongside a flagship model launch signals the category is maturing from research project to sales line item. For a security-conscious buyer, that is a vendor roadmap worth tracking before a competitor's offensive capability forces the decision.
The practical takeaway for EU and UK teams evaluating AI vendors is to treat pricing commitments as short-term, not long-term guarantees, and to start asking vendors now what their defensive product roadmap looks like, rather than after an incident makes the question urgent. Vendors that can name a defensive product this quickly are unlikely to be the last ones to try.
Servola Journal
We do this for everyone trying to keep up with what technology is doing to our lives. The people who build it, and the people it happens to. The Servola Journal exists so that what we learn belongs to all of them.
Nobody pays us for this. No ads, no paywall, free to everyone. We just believe that understanding what's happening to all of us shouldn't depend on who can afford to pay for it.
If it gave you something today, tell us to keep going. Follow us, leave a like, or write a positive comment. We read every one, and they are what keeps us going.
Read next: The First AI Benchmark Neither Side Could See | DeepSeek's Peak Pricing Exposes AI Cost Lock-In
