Free Gemini Drops From Three Models to One on 9 October, and AI Plus Loses Pro

Free Gemini had three models yesterday and one today. Google's own help page, titled Changes to Gemini model access and limits, now carries the sentence that does the work: "These changes will start to take effect for users without an AI subscription on October 9th." When it lands on an account, the model picker in the Gemini app goes from Flash-Lite, Flash and Pro to Flash-Lite alone.

What actually happened

The new matrix is one tick where there were three. Under the heading "Model availability after these changes take effect", Google's change notice shows a single checkmark for a plan with no subscription, two for AI Plus, and three each for AI Pro and AI Ultra. The cheapest paid tier gets an unusual treatment: "For users with an AI Plus subscription, you should receive an email that explains when these changes will take effect for you." So the $4.99 tier loses Pro, but on a date Google has not published for anyone to check.

Free users are losing the two models Google itself calls the capable ones. The separate Gemini Apps limits page defines the lineup: Flash-Lite is "an efficient workhorse model designed for speed", Flash "balances speed and reasoning to solve a large variety of problems", and Pro is "our most advanced model that excels at complex math and coding prompts with high performance and reasoning". The free tier keeps the first and hands the other two to subscribers. That same page, fetched on the morning of 9 October, still prints its older table with "Yes" against all three models for users without a plan — Google's two help documents currently disagree, and only one of them carries the date.

Version numbers and prices are third-party reporting, and are labelled that way here. 9to5Google, which read the updated document on 3 October, identifies the affected models as Gemini 3.5 Flash-Lite, 3.6 Flash and 3.1 Pro, prices AI Plus at $4.99 a month, AI Pro at $19.99 and AI Ultra at $99.99 or $199.99, and reports that AI Pro gains the Deep Think option for "maximum parallel reasoning" that until now sat only on the Ultra plans. Google's own pages publish no prices and no version numbers.

There is a new dial, and it spends budget. The change notice now tells users they can pick an effort level for each available model — "effort level from low, medium, and high" — and warns that more effort "will also use more of your limit". That limit is the compute-based system Google introduced on 17 May 2026: prompts refresh every five hours until a weekly cap is reached, and the work Google lists as expensive — media generation, images, video and music, Deep Research, the Pro model, extended thinking and Deep Think — burns it faster. Allowances remain tiered at standard limits, 2x, 4x, and 5x or 20x of AI Pro depending on the subscription.

The scope is narrow, and the reason is unstated. The notice covers Gemini Apps used with a personal Google account. It says nothing about work or school accounts, the Gemini API, AI Studio, Google Search or AI Overviews, gives no country-by-country schedule — only that changes "start to take effect" on 9 October — and offers no explanation for the decision.

Secondary item of the day: Anthropic pointed its cyber tooling at defenders. On 8 October the company launched the Anthropic Cyber Mission: a Critical Infrastructure Defense Program that brings frontier Claude models, on-site engineers and threat research to eleven founding partners — Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation — plus OSS Scanner, a free opt-in service that gives open-source projects periodic security scans run by Anthropic's strongest models at no cost. Anthropic says six months of scanning found "over 29,000 candidate vulnerabilities" of which roughly 6,000 were manually triaged, and that on the CyberGym benchmark language models went from finding under 20% of vulnerabilities at the start of last year to over 85% this year.

Why this matters for Indian SMBs and mid-market teams

• The downgrade arrives silently. A model swap is not an error message: the prompt still succeeds, the reply still reads like English, and nobody flags that the reasoning behind it got thinner. Teams using free Gemini to summarise invoices, draft replies or parse long PDFs should re-run five of their real prompts today and compare — the difference in quality is the actual cost of this change, and it never shows up on a bill.
• Continuity now has a published price. Keeping Flash costs $4.99 a month and keeping Pro $19.99, per 9to5Google's reading of US pricing (Google's own help pages publish no prices, and Indian pricing is not on them). For a ten-seat team that is roughly $600 a year to hold on to Flash and about $2,400 to hold on to Pro — cheaper than the quiet quality regression it prevents, but it should be a decision you take, not one a help page takes for you.
• Effort level is a budget control, not a slider to leave on high. Every high-effort prompt spends a larger slice of a five-hour compute window. For bulk work — ticket classification, catalogue copy, notification text — low or medium is the right default; high effort belongs on the handful of prompts per day where the answer actually has to be right first time.
• Customer-facing work should not sit on a free model picker. If your website assistant, WhatsApp bot or internal search is quietly backed by free Gemini, a vendor help page just changed what your customers read. Pin the model and version through an API, log which model answered, and keep a second provider wired up so a policy change is a ticket rather than an incident.
• Google's own documents disagree today. One help page dates the change, another still shows the old access — normal during a rollout. Treat the page with the date as authoritative, and re-check your own account's picker rather than any table, including ours.

Where Guruji Corporation fits in

Nobody plans for their AI stack to shrink underneath them because of a help-page update. We run the audit that stops it from being a surprise: an inventory of which model sits behind each workflow you depend on today — free tier, paid tier, or API — measured on your own documents so you have a before-and-after number instead of a feeling; a routing plan that sends summarisation and drafting to Flash-Lite-class models at scale and reserves Pro-class reasoning for the analysis that earns it; and effort-level and usage-limit settings written down as policy, so cost per prompt is predictable and the 5-hour refresh window is a budget you chose.

If your team has been living on free Gemini, this week is the moment to find out what it is actually running on. Talk to us and we will inventory your AI usage, benchmark your real prompts across the Flash-Lite, Flash and Pro tiers plus a second provider, and hand you a migration plan with per-seat costs — before the next notice replaces this one.

Sources - Changes to Gemini model access and limits — Gemini Apps Help - Gemini Apps limits & upgrades for Google AI subscribers - Gemini app limiting what models free & AI Plus users can access — 9to5Google - Introducing the Anthropic Cyber Mission - Launching an opt-in vulnerability-finding service for open-source software — Anthropic