Gemini Pro news
Every thread that mentions this topic, across all sections
Gemini Pro subscribers get 200 monthly Colab credits with RTX Pro 6000 Blackwell GPU access
A Reddit user shared that Google's Gemini Pro subscription includes 200 free Google Colab compute credits monthly, unlocking the Colab G4 GPU (Nvidia RTX Pro 6000 Blackwell Server Edition, 96GB VRAM) for roughly 30+ hours per month. They use it to run Qwen 27B locally with the Cline extension in VS Code at ~50 tokens/second, as a backup when their Antigravity quota runs out.
A Reddit user shared that Google's Gemini Pro subscription includes 200 free Google Colab compute credits monthly, unlocking the Colab G4 GPU (Nvidia RTX Pro 6000 Blackwell Server Edition, 96GB VRAM) for roughly 30+ hours per month. They use it to run Qwen 27B locally with the Cline extension in VS Code at ~50 tokens/second, as a backup when their Antigravity quota runs out.
Google is changing Gemini model availability depending on your subscription plan
Google announced changes to Gemini Apps model availability on personal accounts taking effect October 9, 2026. Free users will be limited to Flash Lite, AI Plus subscribers get Flash added, and AI Pro/Ultra subscribers keep access to Pro plus Deep Think. Users can also select an effort level (low, medium, high) per model, trading answer thoroughness against usage limits.
Google announced changes to Gemini Apps model availability on personal accounts taking effect October 9, 2026. Free users will be limited to Flash Lite, AI Plus subscribers get Flash added, and AI Pro/Ultra subscribers keep access to Pro plus Deep Think. Users can also select an effort level (low, medium, high) per model, trading answer thoroughness against usage limits.
Gemini has two usage counters, and the short one resetting doesn't mean you're back
Reddit users document how Google's Gemini usage limits work and how they are tightening. A detailed post explains Gemini tracks two counters — a 'Current usage' quota refilling every few hours and a 'Weekly limit' resetting weekly — so users who exhaust the weekly cap stay locked out even after the short timer resets. A separate user reports Google has re-implemented usage limits on the Flash-Lite 'thinking' version, which had been free the day before.
Reddit users document how Google's Gemini usage limits work and how they are tightening. A detailed post explains Gemini tracks two counters — a 'Current usage' quota refilling every few hours and a 'Weekly limit' resetting weekly — so users who exhaust the weekly cap stay locked out even after the short timer resets. A separate user reports Google has re-implemented usage limits on the Flash-Lite 'thinking' version, which had been free the day before.
Google limits Gemini Deep Think to Ultra users, walking back October 9 access promised to Pro subscribers
Reddit users spotted Google quietly editing its Gemini support article to remove Deep Think from the Pro tier, walking back previously promised October 9 access, with only Ultra subscribers now listed. Days later, users reported the AI Plus plan also lost access to the pro model, leaving Plus with a flash model similar to the free tier. The changes were made without announcement, frustrating paying subscribers.
Reddit users spotted Google quietly editing its Gemini support article to remove Deep Think from the Pro tier, walking back previously promised October 9 access, with only Ultra subscribers now listed. Days later, users reported the AI Plus plan also lost access to the pro model, leaving Plus with a flash model similar to the free tier. The changes were made without announcement, frustrating paying subscribers.
Gemini Pro model down for users with and without Google AI Pro subscriptions, per Reddit report
Reddit users report Google's Gemini Pro (3.1 Pro) failing en masse starting October 8, 2026, refusing even basic prompts with canned refusals like "I'm just a language model" or "I'm not programmed for that." Flash and Flash-Lite models keep working, and paid and free accounts are both affected. By October 9 the issue persists, with some inputs crashing entirely, while DownDetector shows a broader Gemini outage spike.
Reddit users report Google's Gemini Pro (3.1 Pro) failing en masse starting October 8, 2026, refusing even basic prompts with canned refusals like "I'm just a language model" or "I'm not programmed for that." Flash and Flash-Lite models keep working, and paid and free accounts are both affected. By October 9 the issue persists, with some inputs crashing entirely, while DownDetector shows a broader Gemini outage spike.
PolyRoll – 3D marble rolling game for mobile and desktop
PolyRoll is a 3D marble game inspired by Super Monkey Ball and Marble Blast, playable on mobile and desktop. It was built almost entirely via agentic coding: the creator planned with Gemini Pro and had Claude Sonnet generate a working game in a few hours. It solves nothing beyond entertainment, but demonstrates how quickly playable games can be produced with LLM coding agents.
PolyRoll is a 3D marble game inspired by Super Monkey Ball and Marble Blast, playable on mobile and desktop. It was built almost entirely via agentic coding: the creator planned with Gemini Pro and had Claude Sonnet generate a working game in a few hours. It solves nothing beyond entertainment, but demonstrates how quickly playable games can be produced with LLM coding agents.
Reddit user asks whether Gemini Pro degraded after June–September gap
A Reddit user returning to Gemini Pro after a June-to-September gap reports the model seems much worse and asks whether to subscribe to Claude instead. The post reflects ongoing community concerns about model quality drift but contains no benchmark data or official response.
A Reddit user returning to Gemini Pro after a June-to-September gap reports the model seems much worse and asks whether to subscribe to Claude instead. The post reflects ongoing community concerns about model quality drift but contains no benchmark data or official response.
CERTID Benchmark Shows Reasoning Models Are Accurate but Unsound on Causal Identification
A new paper introduces CERTID, a formal pipeline that certifies whether causal effects are identifiable and verifies model-returned formulas against structural causal models. Evaluations of Gemini Flash, Gemini Pro, and GPT5.5 on 1,200 certified instances show accuracy is a poor proxy for soundness, with false-claim rates on non-identifiable queries varying seventeen-fold across models.
A new paper introduces CERTID, a formal pipeline that certifies whether causal effects are identifiable and verifies model-returned formulas against structural causal models. Evaluations of Gemini Flash, Gemini Pro, and GPT5.5 on 1,200 certified instances show accuracy is a poor proxy for soundness, with false-claim rates on non-identifiable queries varying seventeen-fold across models.
Usuário relata que o Gemini Pro degrada para texto prolixo e carregado de adjetivos após muitos prompts na mesma conversa
A Reddit user reports Gemini Pro progressively degrading during long prompt sessions, adding excessive useless adjectives until text becomes nearly unreadable. The user says this has happened multiple times but found no online discussion. They asked others whether they had experienced the same issue.
A Reddit user reports Gemini Pro progressively degrading during long prompt sessions, adding excessive useless adjectives until text becomes nearly unreadable. The user says this has happened multiple times but found no online discussion. They asked others whether they had experienced the same issue.
Gemini Pro + Canvas says “I’m just a language model” on every request
A Reddit user reports Gemini Pro + Canvas on their main account has malfunctioned since around September 25, 2026: Canvas requests return refusals like "I'm a language model. I can't do that" instead of generating documents. The issue appears account-specific rather than prompt- or browser-specific, and Google Support has escalated the case to a higher-level specialist.
A Reddit user reports Gemini Pro + Canvas on their main account has malfunctioned since around September 25, 2026: Canvas requests return refusals like "I'm a language model. I can't do that" instead of generating documents. The issue appears account-specific rather than prompt- or browser-specific, and Google Support has escalated the case to a higher-level specialist.
Nano Banana 2 on Google Pro subscription generating low-quality images; download full size button removed
A paying Google Pro subscriber reports Nano Banana 2 image quality dropped over two days, producing ~500KB JFIF files at free-tier quality. The 'download full size' option also disappeared despite normal usage limits. The complaint raises quality-of-service concerns for paid image-generation plans.
A paying Google Pro subscriber reports Nano Banana 2 image quality dropped over two days, producing ~500KB JFIF files at free-tier quality. The 'download full size' option also disappeared despite normal usage limits. The complaint raises quality-of-service concerns for paid image-generation plans.
Reddit user finds Gemini Pro falls short of Claude for everyday tasks despite extended thinking
A Reddit user compares Gemini Pro (with extended thinking) against Claude for daily non-coding tasks like health, finance, and productivity, and finds Gemini frustratingly weaker despite reviewer claims that the big three are equal for basic chatting. The post asks whether others see the same gap or whether prompting changes would help. It matters as practitioner feedback on consumer-grade assistant quality beyond benchmarks.
A Reddit user compares Gemini Pro (with extended thinking) against Claude for daily non-coding tasks like health, finance, and productivity, and finds Gemini frustratingly weaker despite reviewer claims that the big three are equal for basic chatting. The post asks whether others see the same gap or whether prompting changes would help. It matters as practitioner feedback on consumer-grade assistant quality beyond benchmarks.
Gemini Pro outage: model rejects all prompts while Flash and Flash-Lite work, Reddit users report
Users report Gemini Pro rejecting even basic prompts with capacity/refusal messages over several hours on 2026-09-11, while Gemini Flash and Flash-Lite work normally. Multiple reports appear within a two-hour window with no stated cause or fix, indicating an apparent service-side incident affecting the Pro model specifically.
Users report Gemini Pro rejecting even basic prompts with capacity/refusal messages over several hours on 2026-09-11, while Gemini Flash and Flash-Lite work normally. Multiple reports appear within a two-hour window with no stated cause or fix, indicating an apparent service-side incident affecting the Pro model specifically.
How do you maintain a reliable external “patient chart” that multiple LLMs can use without stale facts taking over?
A Reddit user building an 'AI Chief of Staff' system asks how to maintain a reliable external 'patient chart' that multiple LLMs can read without stale facts corrupting state. They report that most viral implementations of this pattern collapse because the LLM loses track of 'what is true.'
A Reddit user building an 'AI Chief of Staff' system asks how to maintain a reliable external 'patient chart' that multiple LLMs can read without stale facts corrupting state. They report that most viral implementations of this pattern collapse because the LLM loses track of 'what is true.'
Gemini Pro users report hitting 5-hour limits faster after Gemini 3.8 Flash launch
Gemini Pro subscribers on Reddit report the 5-hour usage window draining far faster than usual since the Gemini 3.8 Flash launch, with lockouts after roughly two hours of normal study and engineering use. The complaints suggest effective limits dropped across both available models, though Google has not confirmed any change.
Gemini Pro subscribers on Reddit report the 5-hour usage window draining far faster than usual since the Gemini 3.8 Flash launch, with lockouts after roughly two hours of normal study and engineering use. The complaints suggest effective limits dropped across both available models, though Google has not confirmed any change.
DeepMind executives spotlight Flash models and orchestration strategy, fueling speculation Gemini Pro is winding down
Google DeepMind executives told CNBC and the G20 Innovation meeting that the recent Gemini Flash models outperformed expectations and will anchor the product strategy going forward. Tulsee Doshi said Flash performance 'really surprised us in positive ways,' and Demis Hassabis described Gemini evolving into a general-purpose layer coordinating cheaper specialized models and agents. The statements suggest the Gemini Pro line may be deprioritized.
Google DeepMind executives told CNBC and the G20 Innovation meeting that the recent Gemini Flash models outperformed expectations and will anchor the product strategy going forward. Tulsee Doshi said Flash performance 'really surprised us in positive ways,' and Demis Hassabis described Gemini evolving into a general-purpose layer coordinating cheaper specialized models and agents. The statements suggest the Gemini Pro line may be deprioritized.
Independent test finds Grok's political bias depends on how you ask
An AI ethics nonprofit tested six frontier models across seven bias datasets, finding Grok's political bias varies with question phrasing. The test covered 20,600 examples and included GPT-5.4, Claude Sonnet 4.6, Claude Opus 4.7, Gemini Pro, Gemini Flash, and Grok 4.3.
An AI ethics nonprofit tested six frontier models across seven bias datasets, finding Grok's political bias varies with question phrasing. The test covered 20,600 examples and included GPT-5.4, Claude Sonnet 4.6, Claude Opus 4.7, Gemini Pro, Gemini Flash, and Grok 4.3.
Students outside US report free Gemini Pro access
Students outside the US are reporting that they can access Gemini Pro for free, according to a Reddit post. The post asks whether this is the case, suggesting a possible expansion of Google's free tier for students.
Students outside the US are reporting that they can access Gemini Pro for free, according to a Reddit post. The post asks whether this is the case, suggesting a possible expansion of Google's free tier for students.
Users report Gemini Pro per-prompt quota cost rises with heavy usage, resets after idle
A Reddit user reports that per-prompt quota cost for Gemini Pro (3.7 Flash Extended) varies with recent usage: heavy continuous use increases the percentage cost from ~1% to 5%+, while leaving an account idle for a few days resets it. The observation suggests dynamic quota pricing based on usage patterns.
A Reddit user reports that per-prompt quota cost for Gemini Pro (3.7 Flash Extended) varies with recent usage: heavy continuous use increases the percentage cost from ~1% to 5%+, while leaving an account idle for a few days resets it. The observation suggests dynamic quota pricing based on usage patterns.
Gemini Enterprise Business users report app shows 3.6 Flash despite 3.7 Flash upgrade
A Gemini Enterprise Business user upgraded and was told by the app they are on 3.7 Flash, but the model dropdown still displays 3.6 Flash. The user finds the discrepancy disconcerting and is seeking clarification on which model is actually active.
A Gemini Enterprise Business user upgraded and was told by the app they are on 3.7 Flash, but the model dropdown still displays 3.6 Flash. The user finds the discrepancy disconcerting and is seeking clarification on which model is actually active.
Users test subscription limits of Gemini Pro, Ultra, Claude Pro, and Opencode Go
A Reddit post reports on testing the subscription limits of several AI services, including Gemini Pro, Gemini Ultra, Claude Pro, and Opencode Go. The tests reveal how many requests or usage each plan allows before hitting limits, providing practical information for users comparing these subscriptions.
A Reddit post reports on testing the subscription limits of several AI services, including Gemini Pro, Gemini Ultra, Claude Pro, and Opencode Go. The tests reveal how many requests or usage each plan allows before hitting limits, providing practical information for users comparing these subscriptions.
Users weigh switching from Claude Pro to Gemini Pro for research-heavy use
A Reddit user considering a switch from Claude Pro to Gemini Pro for research-heavy study help asks the community for input, citing high token costs from running an AI agent via OpenRouter. The thread gathers practitioner opinions on Gemini's browser-based research capabilities versus Claude's strengths.
A Reddit user considering a switch from Claude Pro to Gemini Pro for research-heavy study help asks the community for input, citing high token costs from running an AI agent via OpenRouter. The thread gathers practitioner opinions on Gemini's browser-based research capabilities versus Claude's strengths.
Users report Gemini Pro degraded for teaching, seek alternatives
A user reports that Gemini Pro has become noticeably worse at teaching new technologies, despite still being good at coding. They are seeking alternative models that match its former explanatory quality, noting Flash 3.7 lacks the teaching ability.
A user reports that Gemini Pro has become noticeably worse at teaching new technologies, despite still being good at coding. They are seeking alternative models that match its former explanatory quality, noting Flash 3.7 lacks the teaching ability.
Users compare GPT Pro, Fable, Opus, GPT Extra Hard, Gemini Pro on research critique
A researcher testing agent memory integrity fed study results to five top AI models and asked each to critique the work. The user reports all models performed well, but GPT Pro stood out for the most thorough system thinking across breadth and depth.
A researcher testing agent memory integrity fed study results to five top AI models and asked each to critique the work. The user reports all models performed well, but GPT Pro stood out for the most thorough system thinking across breadth and depth.
Gemini Pro subscribers report widespread generation errors and generic refusals
Gemini Pro subscribers are experiencing constant generation errors starting August 19, 2026, with simple prompts failing and returning generic refusals like "having a hard time fulfilling the request." The issue occurs even in new chat sessions and on the first turn, affecting tasks like parsing network IP data and structuring tables.
Gemini Pro subscribers are experiencing constant generation errors starting August 19, 2026, with simple prompts failing and returning generic refusals like "having a hard time fulfilling the request." The issue occurs even in new chat sessions and on the first turn, affecting tasks like parsing network IP data and structuring tables.
Jio's 18-month Gemini Pro offer: eligibility and pitfalls explained
A Reddit post clarifies that the '18-month Gemini Pro' offer is Jio's official India promotion, requiring users aged 18+ with a Jio SIM and an active unlimited 5G plan of ₹349 or more. The claim is done via MyJio, tied to a Gmail account, and each eligible number can claim once, with the plan needing to stay active.
A Reddit post clarifies that the '18-month Gemini Pro' offer is Jio's official India promotion, requiring users aged 18+ with a Jio SIM and an active unlimited 5G plan of ₹349 or more. The claim is done via MyJio, tied to a Gmail account, and each eligible number can claim once, with the plan needing to stay active.
Users report Gemini degradation: errors, hallucinations, and instruction failures
Multiple users report a recent degradation in Gemini's reliability, including errors, hallucinations, and failure to follow instructions, particularly in document-processing tasks. The issues appear to have started within the past week, affecting long-time users who previously had few problems.
Multiple users report a recent degradation in Gemini's reliability, including errors, hallucinations, and failure to follow instructions, particularly in document-processing tasks. The issues appear to have started within the past week, affecting long-time users who previously had few problems.
Student in Africa seeks cheapest Claude Pro access as Gemini student deal ends
A 4th-year mechatronics student in Africa asks for the cheapest way to get Claude Pro, as their free year of Gemini Pro for students is ending and the $25/month cost is unaffordable. They need higher limits for academic thesis work and coding, and want to use Claude Code.
A 4th-year mechatronics student in Africa asks for the cheapest way to get Claude Pro, as their free year of Gemini Pro for students is ending and the $25/month cost is unaffordable. They need higher limits for academic thesis work and coding, and want to use Claude Code.
Users question Gemini Pro's lack of third-party tool support for vibe coding
A Reddit user asks whether Gemini Pro's inability to connect to third-party tools like Pi, Opencode, and Hermes disqualifies it for vibe coding, noting that only Antigravity can be run on a VPS and called from another AI agent. The post highlights a limitation in Gemini's subscription model for developers who prefer flexible tool integration.
A Reddit user asks whether Gemini Pro's inability to connect to third-party tools like Pi, Opencode, and Hermes disqualifies it for vibe coding, noting that only Antigravity can be run on a VPS and called from another AI agent. The post highlights a limitation in Gemini's subscription model for developers who prefer flexible tool integration.
Gemini Pro users report Spark beta appearing and disappearing in app
Gemini Pro subscribers report that the Spark feature intermittently appears as a beta in the app and then disappears without warning. The unpredictable visibility suggests an unstable rollout or A/B testing by Google.
Gemini Pro subscribers report that the Spark feature intermittently appears as a beta in the app and then disappears without warning. The unpredictable visibility suggests an unstable rollout or A/B testing by Google.