Ahead of Gemini 3.0 release, Gemini 2.5 Pro users report degraded performance

Paying subscribers are fuming. With Gemini 3.0 potentially dropping around October 9 (might be later), users across Reddit have noticed something wrong with Gemini 2.5 Pro over the past two weeks. What was once Google’s flagship AI model now struggles with basic tasks, spits out hallucinations, and times out on routine requests.

One user described how their Gem started inventing dramatic events that had no basis in referenced documents. “Made up events, dramatic stuff that clearly has no basis in the referenced doc,” they wrote, adding that clickable citation links mysteriously vanished.

The complaint struck a nerve, with others reporting similar experiences. Someone asked about their MacBook Pro M4’s lifespan only for Gemini to claim the device hadn’t been released yet. Another got information about a shoe brand, then watched the AI claim that same brand didn’t exist when asked a follow-up question.

Another thread from a long-time user captured the frustration more bluntly. “I’ve used it to great success for many months, but past week I has started to time out about 50% of my requests,” they wrote, noting that code generation became so bad they cancelled their subscription and went back to coding manually. That post gathered over 230 upvotes, with dozens of users confirming they faced identical problems.

Some users noticed that Gemini 2.5 Flash, the supposedly lighter model, started outperforming Pro. One person tried summarizing a YouTube video and got told they were attempting a Rick Roll. System administrators who relied on Pro to analyze error logs now get condescending explanations instead of technical analysis. “It talks to me like I’ve pumped the information into myfirsterrormessage.com,” one wrote.

Additional complaints kept piling up across multiple subreddits, and forum discussions showed the same story everywhere. The most popular theory? Google is training Gemini 3.0 and has starved 2.5 Pro of computational resources. Some users noticed better performance late at night when fewer people use the service. Others think Google deployed a quantized, stripped-down version to free up TPUs for the upcoming model. But of course, this is nothing more than plain speculation.

Google’s silence isn’t helping. No official acknowledgment, no explanation, just increasingly frustrated users questioning if their subscriptions are worth keeping.

That said, the hype for Gemini 3.0 is already reaching the ceiling, with leaks suggesting that users can expect big leaps in performance. So maybe, just maybe, hang in there and don’t call it quits on your Gemini AI subscription just yet.