What Google Shipped on September 2, and Where It Runs
Google released Gemini 3.8 Flash and a second variant called Gemini 3.8 Flash Cyber on September 2, 2026, and made both available the same day across AI Pro, AI Ultra, AI Mode, Sheets, Antigravity, AI Studio, and the API.
The company confirmed the launch on its own blog, and outlets including 9to5Google and The Register verified the same day that pricing carried over unchanged from the prior release: $0.75 per million input tokens and $3.75 per million output tokens.
A Three-Week Cadence, Not a Three-Month One
Gemini 3.8 Flash is the third Flash-tier model Google has shipped in roughly three months, arriving after 3.7 Flash and other recent Flash updates on a cadence that 9to5Google and The Register both clocked at about three weeks.
| Gemini 3.8 Flash input price | $0.75 per million tokens (unchanged) |
|---|---|
| Gemini 3.8 Flash output price | $3.75 per million tokens (unchanged) |
| Flash-tier releases in ~3 months | 3rd (after 3.7 Flash and other Flash updates) |
| Measured cadence since 3.7 Flash | about 3 weeks |
| Cyber variant correct-patch advantage | 2.6x versus larger commercial rival models |
Those numbers describe a company iterating on its cheapest, highest-volume tier faster than most enterprise buyers refresh a vendor scorecard, let alone a signed contract.
Flat Pricing on a Faster Clock Commoditizes the Mid-Tier
Google held Gemini 3.8 Flash at $0.75 per million input tokens and $3.75 per million output tokens, about EUR 0.69 and EUR 3.46 at current exchange rates, exactly what it charged for the model it replaced.
That combination matters more than a spec bump for an EU buyer: it is Google actively commoditizing the mid-tier inference layer that most EU small and mid-market enterprises actually build production apps on, not the flagship tier reserved for frontier demos. A team that shipped an app in July on what was then the current Flash model is, by September, running it on a tier Google itself now treats as superseded, at a price that never moved.
What the Three-Week Clock Does to an Annual Vendor Contract
A European enterprise that signed an annual inference agreement in July negotiated a rate against a model Google effectively retired within weeks, before the contract's ink fully dried.
Procurement cycles built around quarterly or annual review dates cannot track a vendor releasing three comparable models in three months, so the negotiated discount that looked competitive at signing stops looking like a deal almost immediately, and the model-selection decision embedded in that contract ages out faster than most procurement teams can convene a review.
The practical fix is to write repricing and reselection clauses into any multi-year inference agreement now, rather than treating a signed rate as fixed for the term, because Google has just demonstrated it will keep shipping equivalent-tier models faster than most vendor-management calendars turn.
The Cyber Variant's Vulnerability-Patch Claim, in a Crowded Week for AI Security
Google's Chrome Security team reported that Gemini 3.8 Flash Cyber produced 2.6 times more correct vulnerability patches than larger commercial rival models in its internal testing, a figure Google published alongside the launch rather than as a separate research paper.
Access to the Cyber variant runs through a new gate Google calls the Fairwind Program, not the open channel the base Gemini 3.8 Flash uses, which keeps the vulnerability-patching capability away from general availability even as the base model ships everywhere at once. EU security teams are watching more than one vendor race to claim an edge in AI-assisted vulnerability work this same week, and a gated access program is itself a signal worth reading: Google is not yet confident enough in the claim, or the misuse risk, to ship it the way it ships everything else.
What EU Buyers Should Actually Do With This Release
EU buyers should treat Gemini 3.8 Flash as a scheduling problem, not a one-time upgrade decision, given the cadence Google has now shown three times in three months.
Build model-reselection checkpoints into procurement cycles at least quarterly rather than annually, insist on repricing or re-benchmarking clauses in any inference contract signed today, and budget engineering time to swap model versions the way a cloud team budgets time for patching, because the alternative is running production traffic on a tier the vendor itself has already moved past.
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: Gemini's Agentic Video Cuts Token Costs 88% | 800,000 Reservations Did Not Buy A Launch
