Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

Google Moves Gemini Free Users to Flash-Lite From 9 October

Google will limit users without a paid Gemini plan to the Flash-Lite model from 9 October, according to a support document the company updated this week. Free users lose 3.6 Flash and 3.1 Pro, and $4.99 per month AI Plus subscribers lose Pro.

AI & modelsNewsGrace OkonkwoPublished: 4 October 20264 min readSources 10
Google Moves Gemini Free Users to Flash-Lite From 9 October

Google confirmed the change in an updated support document this week. 9to5Google reported it on 3 October. Users who pay nothing will be left with the 3.5 Flash-Lite model only. AI Plus subscribers pay $4.99 per month. They keep Flash-Lite and Flash, but lose Pro.

The tiers now line up with a paywall. AI Pro at $19.99 per month and AI Ultra at $99.99 or $199.99 per month keep access to all three models, according to the table published by The Decoder. AI Plus customers will "receive an email that explains when these changes will take effect," 9to5Google reported, quoting the support page. That is a downgrade from the current free tier, which offers 3.6 Flash and varying access to 3.1 Pro.

For most people the practical effect is small, at least at first. Casual users rarely know which model answers their prompt. The ones who do tend to pay already. The bigger question is what Google does next with Gemini 4 Argon. 9to5Google reported that Argon will first go to AI Ultra subscribers. The company has not said whether it counts as a Pro model or sits in a tier above it.

Pro users get Deep Think, but the top model stays unclear

There is one upgrade in the same document. AI Pro subscribers will get the Deep Think option for "maximum parallel reasoning," 9to5Google reported. Until now that feature has been limited to the $99.99 and $199.99 per month Ultra plans. The Gemini app also looks set to add "low," "medium" and "high" effort levels for each model in October. That would raise answer quality at the cost of using more of a user's limit.

Google's move copies OpenAI's freemium split. The free tier gets the cheapest model and paying customers get the rest. The Decoder noted that the change may also prepare the ground for Gemini 4 Argon, which could cost more to run than the current models. Neither company has published per-query costs for these tiers.

The timing sits awkwardly against Google's other AI work. Gemini 4 Carbon, a model aimed at finding and fixing security flaws, has been the subject of recent coverage. The Argon launch is expected to draw the most attention from developers. Whether Argon lands inside the Pro tier or above it will decide how much of the paid base actually gets the newest model.

Elsewhere in models, the benchmark picture is thin

Model releases have kept coming through the autumn. Mistral's Le Chonk model was made available for download in late October. Aleph Alpha published Kolibri-1, a 78B open-weight model with 1M context, under Apache 2.0 on Hugging Face. Independent comparison work has been less generous. SemiAnalysis's GDPx benchmark, reported on 1 November, put the best AI model at 12% on its tasks. Separate coverage on 3 November said finance and legal tasks score zero.

Smaller, narrower models are also arriving. Ideogram released version 4.5 of its image model on 4 October. It has a precise edit mode that the company says avoids artifact buildup across repeated edits. The company prices a high-quality edit at $0.22 per image and a low-quality one at $0.03, with no subscription required. Ollama 0.35.1 added Cloudflare's Clef and Clef Flash, two classifier models that return yes-or-no scores rather than text.

Jerry Gamblin, who serves on the CVE Program's Quality, Automation and Consumer working groups, ran Clef Flash over all 27,489 CVEs published in the 60 days to 3 October. Outside the Linux kernel, 989 of 23,736 descriptions, or 4.2%, never state what an attacker could do, he wrote on 4 October. The kernel's own policy leaves impact to the user. 3,661 of its 3,753 records state none, taking the all-CVE figure to 4,650 of 27,489, or 16.9%. He describes the working-group role as a disclosure, not a CVE Program position.

Anthropic, meanwhile, has started asking Claude users to share voice conversations for model training. BleepingComputer reported on 4 October that the prompt is optional and off by default. It is separate from the existing setting for chats and coding sessions. Users can turn it off or delete the data in settings.

None of this settles the question Google's support page leaves open: what happens to the free tier when Argon ships. The company has not said whether Flash-Lite users will ever see it.

Comments 0

Sources

10
  1. 01Gemini app limiting what models free and AI Plus users can accessEN
  2. 02Google's new Gemini tiers cut free users to its weakest model and lock $5/month subscribers out of ProEN
  3. 03Anthropic asks Claude users to share voice data for AI model trainingEN
  4. 04Decision Models vs. CVE DataEN
  5. 05Ideogram 4.5: The most precise edit modelEN
  6. 06No Model Required: Text Entropy Rate Filtering Mitigates Iterative Fine-TuningEN
  7. 07OpenAI's latest features take direct aim at the app store modelEN
  8. 08Language Model "Shape"EN
  9. 09China's space plane appears to have released a mystery object in orbitEN
  10. 10FF7 Revelation director says he "did his best" to fight for a full disc releaseEN

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Grace Okonkwo

Grace Okonkwo

AI, models and technology

Grace Okonkwo covers AI, models and technology for FLASH24, working from primary sources such as model cards, API documentation and benchmark papers rather than vendor summaries. She checks training data provenance, evaluation conditions and reported scores against the underlying datasets before any figure reaches print. She interviews researchers and engineers directly, tracks release calendars from major labs, and compares successive model versions on the same tests. Her own self-hosting, home-network and documentation-reading habits feed straight into that desk, since she tests tools on her own hardware first. She does not publish benchmark claims without a reproducible method.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.