Gemini 3.8 Flash, Workspace and your HubSpot data

Last checked: 12 Sep 2026. Every price, limit and residency statement below was read from Google's and HubSpot's own pages on that date. Google's published price for this model doubles on 1 Jan 2027, so treat anything you read after December as stale until you re-check it.

Google released Gemini 3.8 Flash on 2 Sep 2026. If your company already pays for Google Workspace, the reasonable question is not whether the model is good. It is whether you now have something you did not have last month, what it costs, and what happens to your CRM records if you point it at HubSpot. This post answers those three, and says plainly where Google and HubSpot do not answer them.

Prices are converted from US dollars at £1 = 1.35 US dollars (market rate, 11 Sep 2026). Vendors bill in dollars and UK VAT applies on top.

What actually launched on 2 Sep 2026

Google's launch post describes 3.8 Flash as its workhorse tier, sitting below Pro and above Flash-Lite, and it is the third Flash release in six weeks after 3.5 and 3.7. It shipped alongside a restricted sibling, Gemini 3.8 Flash Cyber, which is access-gated through Google's Fairwind Program for government bodies, critical-infrastructure operators and software maintainers. That one is not available to a normal commercial buyer, so ignore it.

The more useful statement comes from Google itself. The DeepMind model card says "Gemini 3.8 Flash is based on Gemini 3.7 Flash", and that the model "does not have meaningful new capabilities or material increases in performance with respect to the domains outlined in our Frontier Safety Framework compared to Gemini 3.7 Flash". A vendor rarely writes that down. It is worth taking at face value: this is an incremental release on an existing base, not a new generation, and it should not trigger a review of your AI strategy on its own.

The specification and the price

The API model identifier is gemini-3.8-flash. Figures below come from Google's launch post and the Gemini API model listing.

The introductory price is unchanged from 3.7 Flash, and the standard price is double it. If you are building a business case on the current rate, put the January step in the model now. A workload that is comfortable at 56p per million input tokens is a different conversation at £1.11.

Where your data goes, and the part UK buyers should read twice

This is the fact that decides the matter for a lot of UK companies, and it comes from Google Cloud's own documentation, on the data residency page for Gemini Enterprise, updated 3 Sep 2026.

Gemini 3.8 Flash supports at-rest data residency and machine-learning processing in Google's US and EU multi-regions. It does not support the in-country regions. For those, including the United Kingdom (europe-west2), Google's page states the model is "only available in the global region", and an administrator has to confirm a warning dialog to route traffic to the global endpoint.

Gemini 3.5 Flash, by contrast, is supported for both at-rest residency and processing in the in-country regions, the UK included. So if your organisation has committed, in a policy document or to a customer, that personal data stays in the United Kingdom, the newest Flash model is not the one that keeps that promise and an older one is. That is an uncomfortable trade to put in front of a board, but it is the actual position and it is better raised before someone turns the toggle on.

One gap worth naming: neither the model card nor the residency page states Google's position on training on enterprise customer data for this specific model. Do not assume the standard enterprise commitment carries across unchanged. Ask Google to confirm it in writing against the model and the surface you intend to use.

The HubSpot connector for Gemini

There is a real, named integration: the HubSpot connector for Gemini, documented in HubSpot's knowledge base and last updated 16 Jul 2026. It is in public beta. What it does, in HubSpot's own terms:

• It brings HubSpot CRM data into Gemini inside Workspace apps, called with @HubSpot.
• The scope is contacts, companies, deals, their associations and default properties.
• It is read-only. You cannot edit HubSpot data through Gemini.
• It respects HubSpot permissions: a user sees in Gemini only the CRM data they are allowed to see in HubSpot.
• It excludes custom Sensitive Data properties and content from other Sensitive Data covered services, including personal health information.
• It is available across HubSpot products and plans, but requires a paid Workspace Business Standard, Business Plus or Enterprise plan with Gemini Alpha enabled.
• English only.

Two things this connector is not. It is not evidence that Gemini 3.8 Flash powers it: HubSpot's article predates the 2 Sep launch and names no model version, so anyone telling you your CRM is now being read by 3.8 Flash is guessing. And it is not HubSpot's own AI layer. Agent Hub, formerly Breeze Agents, is a separate thing that runs inside HubSpot, and we cover it on AI for HubSpot and on our HubSpot AI page.

What you get for what you already pay

The recurring confusion is whether any of this is included in the Workspace bill you already have. Google's pages are specific in some places and silent in others, so the table marks the silences rather than filling them in.

When 3.8 Flash is the sensible default, and when it is not

Take it as the default when the work is high volume and cost sensitive, the accuracy bar is met by a workhorse model rather than a frontier one, and your data can sit in the EU or US multi-regions. Classifying inbound messages, summarising long documents, first-pass extraction across a large archive: these are the shapes where a price of about 56p per million input tokens changes what is worth automating at all. Take it as the default, too, if what you actually want is for a salesperson to ask a question about a deal without leaving Gmail or Docs, and reading the record is enough.

Do not take it as the default in three cases. First, if UK in-country residency is a contractual or regulatory commitment, because Google's documentation says this model cannot meet it and an older Flash can. Second, if the job involves writing back to the CRM, updating a deal stage, creating a task, logging an outcome, because the connector reads and does not write, and anything that writes is a piece of software somebody has to build and govern. Third, if the decision the model makes carries real consequence, in which case the model tier is the smaller question and the approval step, the audit trail and the scoped credentials are the larger one.

Worth saying clearly: none of this makes Gemini the right or wrong choice against the assistants you may already pay for. Google's own model card is the strongest argument for not rushing. The model choice should follow the workload, the residency position and what the data is allowed to touch, and it should be revisited when those change rather than when a vendor ships a point release. That is how we run it on AI implementation engagements, and it is why the first question is usually about your data rather than about the model.

Questions to put to Google and to HubSpot

Ask Google, in writing, against the specific model and surface you plan to use:

• Confirm whether our Workspace plan includes Gemini 3.8 Flash, or whether an AI Pro or Ultra subscription is required per user.
• Confirm the residency position: our prompts and any HubSpot data in them will be processed in the global region, not in europe-west2. What is the at-rest position for each surface we would use?
• State your commitment on training on our data for this model, on each surface, in the contract rather than in a help article.
• What is the retention period for prompts and responses, and how do we delete them?
• Confirm the price we will pay from 1 Jan 2027 and whether the introductory rate is held for existing customers.

Ask HubSpot:

• Which Gemini model version serves the connector today, and will we be told when it changes?
• Confirm that the connector remains read-only, and tell us what the roadmap is for write access.
• Which of our properties are classed as Sensitive Data, and therefore excluded?
• How does the connector behave for a user whose HubSpot permissions change or who is deactivated?
• Where does connector activity appear in the audit log?

FAQ

Is Gemini 3.8 Flash included in my Google Workspace subscription?
Google's launch post names the Gemini app for Google AI Pro and Ultra subscribers, AI Mode in Search and Gemini in Sheets. It does not state that a base paid Workspace seat includes the model, so treat that as unconfirmed and ask Google against your own plan.

Can Gemini read our HubSpot data?
Yes, through the HubSpot connector for Gemini, called with @HubSpot. It covers contacts, companies, deals, their associations and default properties, it respects each user's HubSpot permissions, and it is read-only.

Does Gemini 3.8 Flash keep our data in the UK?
No. Google Cloud's data residency documentation, updated 3 Sep 2026, states the model is only available in the global region for in-country regions including the United Kingdom (europe-west2). It does support at-rest residency and processing in the EU and US multi-regions. Gemini 3.5 Flash supports UK in-country residency.

What does Gemini 3.8 Flash cost?
Google's published price to 31 Dec 2026 is about £0.56 per million input tokens and about £2.78 per million output tokens, doubling from 1 Jan 2027 to about £1.11 and about £5.56. Converted from US dollars at £1 = 1.35 US dollars on 11 Sep 2026, with UK VAT on top.

Is Gemini 3.8 Flash a big upgrade on 3.7 Flash?
Google's own model card says it is based on Gemini 3.7 Flash and does not have meaningful new capabilities or material increases in performance against the Frontier Safety Framework domains compared with 3.7 Flash.

Does the HubSpot connector run on Gemini 3.8 Flash?
HubSpot does not say. Its knowledge-base article was last updated 16 Jul 2026, before the 2 Sep 2026 launch, and names no model version.

Where to go next

If you are weighing Gemini against what you already pay for, the useful work is not a benchmark comparison. It is establishing which of your data can leave the UK, which decisions you are prepared to let software make, and what has to be true before any of it is worth building. Start with a diagnostic, or Request a Quote if you already know what you want built.

SpotDev is an AI and digital transformation consultancy that builds real software, an OpenAI Select Partner and a Claude Registered Partner.