Gemini 3 goes mainstream: hosts test it as a ChatGPT replacement and find the model impressive but the app buggy

Summary

Google's Gemini 3 has moved into direct competition with ChatGPT. An observer with deep foundation model relationships said after two hours that he's used ChatGPT daily for three years and won't go back, citing leaps in reasoning, speed, image handling, and video as "insane."

Daily testing on iOS revealed an impressive underlying model paired with a buggy app. Heavy users spending roughly 15 minutes per day noticed Gemini 3 sizes responses more intelligently than ChatGPT: a one-paragraph answer gets one paragraph instead of a two-page dissertation. The model also feels faster, though some of that perception reflects comparison fatigue from waiting 10 minutes for GPT-4o Pro responses. When Gemini searches the web, it says "we're searching Google," which triggers psychological trust built over 25 years even though the underlying web is identical.

Gemini's image generation model handles complex images and text well enough to summarize Anthropic's model architectures into clean infographics. Multimodal understanding feels richer when images accompany text, more educational than walls of text. Yet the multimodal experience itself fragments.

UI-level problems undermine the strength of the underlying model. Switching between image and text modes requires explicit buttons instead of seamless context switching. Voice input transcription cuts off mid-pause, unlike ChatGPT's two-button system. When one user asked Gemini to generate an image of the most annoying LinkedIn profile, it interpreted the request as a text summary. The app has disconnection errors, prompt submission hangs, and mode-confusion bugs. By the standard that "the app should be flawless," Gemini 3 fails as a product despite model strength.

ChatGPT has accumulated personalization layers over the past year: memory, custom instructions, linked integrations, auth syncing across services. Neither Gemini nor ChatGPT felt truly personalized in practice. Google could theoretically offer far deeper personalization by tapping Gmail, search history, and browsing data. The network effects often cited as ChatGPT's moat appear bolted on rather than organic. Chat history doesn't drive content discovery or social loops. Content quality on day one of Gemini matched or exceeded ChatGPT because both are AI-generated. This suggests the consumer AI market may not be pure winner-take-all but instead a duopoly or weaker oligopoly.

Google is pushing student adoption with one year of Gemini Pro free, betting on workflow lock-in early. OpenAI is bundling advanced thinking queries into its Atlas browser and upping limits for early adopters. The real competitive lever may be monetization timing. Google could run display ads across its free user base tomorrow, but doing so too early risks the app being labeled "the ad one." The approach is to perfect the product and win share first, ads later.

Alphabet stock jumped 6 percent following Gemini 3's launch. Barron's published a piece titled "Buy Google Stock," calling Alphabet "the clear AI winner," a narrative reversal from earlier in the year when the company was described as the AI loser. The market signal is that Google has closed the gap with OpenAI in consumer AI, if not yet overtaken it.