Google has moved Gemini to explicitly tiered access: per community reports quoting Google's own support page and service update, from March 25, 2026 Gemini Pro models are accessible only via paid subscriptions and free-tier users are limited to Gemini Flash. The tiering caps a longer retrenchment traceable across the supplied coverage: after Gemini 3's November 18, 2025 launch, demand-driven 'Basic access' rationing with fluid daily caps replaced predictable free limits in the consumer app (Android Police, BGR, Gadget Hacks), while API free quotas were reportedly cut ~92% — down to ~20 requests/day on Flash — with no advance notice, drawing 'bait and switch' accusations and migration talk on Google's developer forum. Press framing attributes the cuts to capacity-plus-abuse pressures and reads them as deliberate freemium conversion: the free tier becomes a discovery mechanism rather than a durable option for heavy users. Caveats: Google's support page appears here only secondhand via Reddit/press quotes, the numeric caps are user-reported and described as fluctuating, and the sources mix distinct surfaces (consumer app vs. developer API), so exact terms cannot be verified from the supplied material.
Google is enacting, at frontier scale, the hosted-access retrenchment Scott's canon already hedges against — vendor-lock-in, model-perishability and sovereign-software-assurance all price exactly this failure (access repriced or withdrawn underfoot with no tested exit path), and a free tier demoted to a bounded discovery gift is New Freemium doctrine confirmed by the supplier itself. It also bears on what he actively runs: cost-tiered LLM routing with an OpenRouter free primary and free-quota agent/wiki-maintenance jobs are precisely what a ~92% Gemini free-API cut stresses, strengthening the Ollama/gamepc local hedge and handing him dated receipts — to the extent the secondhand support-page terms hold.
ip:concept.new-freemiumip:concept.vendor-lock-inip:concept.model-perishabilityip:framework.sovereign-software-assurancedev:concept.cost-tiered-llm-routingdev:technology.openrouterdev:technology.ollamadev:project.gamepcradar:concept.usage-limitsradar:concept.ai-monetizationradar:concept.inference-economicsradar:gemini-25-october-migration-gapradar:zed-paid-edit-predictionsradar:openai-chatgpt-ads-global-rolloutradar:concept.local-inference
queries asked of Scott's wikis
- inference economics: cost of serving free-tier users at frontier labs
- open-weight / local inference as hedge when hosted API access retrenches
- LLM provider lock-in and switching costs for API-dependent projects
- rate limit and quota backoff handling in agent harnesses
- freemium conversion funnel and tiering patterns in AI products
- free-tier LLM quotas for agent memory and wiki-maintenance jobs
now 0 pts/hpeak 119 pts/hcomments 0/hpeers p14momentum: steady3 platformsage 207h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
2026-10-08T00:35:48Z
Nothing substantive moved — deltas are ±1 on static threads and the recurring velocity_spike flags are the same aging-cohort p90 artifact (143–149 vs p90 132 on a post now parked at 147), so the magnitude-valve 'loud spread' is a stale echo of the hour-8 footprint with zero new evidence, communities, or Google response in four days; rate is 0/h vs ~119 peak. The one real shift is temporal: the countdown entered its final ~24h, so medium heat is event-driven — the consequential phase (does the cut take effect as described, does AI Studio/API free access survive, does Google move) begins Oct 9 and warrants a look at cutover fallout, not engagement.
2026-10-05T09:39:35Z
Nothing in the case's meaning moved: the deltas are +2/+1/+1 points-and-comments on aging threads, the velocity_spike and 'accelerating' momentum flags are decay artifacts against a ~2/h baseline (vs ~119 peak, percentile flattered by a weak aged cohort), and the lone new comment (a paying Pro user still stuck on 3.6 Flash) is unverified testimonial, not corroboration. The case remains a quiet, press-confirmed countdown to the Oct 9 cutover with Google silent; it will re-fire on real cutover evidence, so no material flag is warranted.
2026-10-04T17:06:42Z
First press corroboration — 9to5Google on free AND AI Plus model limits — firms the tiering beyond secondhand screenshots, but the periphery has closed: the report found no audience (1/0), rate is ~2.5/h vs ~119 peak, and the magnitude-valve spread is the same three-platform footprint already alerted at hour 8, so heat drops to low. The case is now a quiet, press-confirmed countdown to the Oct 9 cutover; the sensor's 'accelerating' momentum label is a small-baseline artifact of minor upticks on aging threads (51→59 HN, 10→13 dup) against a 2.5/h overall rate, and the velocity_spike flags are peak-decay noise. Material flag set so polling survives to the dated event.
2026-10-04T16:42:23Z
evidence attached: hn.story.49955364 — Independent 9to5Google report on Gemini model limits for free and AI Plus tiers corroborates the free-access retrenchment episode beyond the original Reddit reports.
2026-10-03T21:00:53Z
Fourth same-echo discussion: an independent Reddit thread pins the cut to a concrete Oct 9, 2026 effective date — superseding the grounding's garbled 'March 25' date — and surfaces unanswered scope questions (AI Studio, student plans), but adds no new terms and found only a modest audience (19/9). That is consolidation, not expansion: the episode is now a dated countdown to the Oct 9 cutover with Google silent. The velocity_spike dirty flag reads as residual noise from the hour-8 peak (current rate ~11/h vs ~113/h peak, cooling); the magnitude-valve spread is the same 3-platform footprint already priced at the hour-8 alert, with no new outlets or communities since, so corroborated/medium stands.
2026-10-03T20:26:17Z
evidence attached: reddit.post.1wwwe6m — Independent Reddit echo of Google's Oct 9 access change corroborates the free-tier-cut episode's spread.
2026-10-03T17:46:31Z
The expansion phase is over: velocity fell from its ~65 pts/h peak to ~19/h by hour 17, the second HN submission of the same support page died at 4 pts/0 comments, and the main HN thread's growth (46→51 pts, 38→45 comments) is amplification of terms already recorded, not new information — no Google response, no new communities or outlets since the alert. The case drops from accelerating to corroborated: a community-confirmed policy change now waiting on external resolution rather than still spreading. Magnitude valve noted — the 3-platform top-decile spread was what earned the high-heat alert at hour 8; with momentum at ~30% of peak and the newest evidence finding no audience, high's within-hours re-look would buy nothing, so medium is the honest delivery price.
2026-10-03T17:23:53Z
evidence attached: hn.story.49945630 — Same Google support-page access/limits story surfacing on HN — cross-community spread evidence for the free-tier retrenchment episode.
2026-10-03T11:28:37Z
The episode crossed from a single Reddit report to top-decile cross-platform spread (Reddit velocity spike, HN pickup climbing 8→27 points within hours), and the HN thread firmed the policy terms: free tier reportedly drops to Flash-Lite only — the same model behind AI Overviews and the Android app — a deeper cut than the grounding's 'limited to Flash'. That makes the free-tier-as-bounded-discovery-gift reading concrete; the live question shifts to Google's response and whether AI Studio/API free access survives the tiering.
2026-10-03T09:25:36Z
evidence attached: hn.story.49942592 — The free-tier cut story crossing from Reddit to HN is independent spread of the same episode, strengthening the retrenchment signal.
2026-10-03T03:45:04Z
grounded: converges/high — Google is enacting, at frontier scale, the hosted-access retrenchment Scott's canon already hedges against — vendor-lock-in, model-perishability and sovereign-s
2026-10-03T03:36:42Z
case created — Concrete, verifiable product-policy change at a frontier provider bearing directly on model-access economics, not covered by any open case.