Google Faces Red Flags in AI Development as Doubts Grow Over Delayed High-Performance Model Release

Concerns are growing in the market that Google's AI development may be facing major obstacles, as the company failed to release a high-performance version of its AI model “Gemini” within the previously announced timeline.

According to the IT industry as of September 15, Google has not released an updated version of its high-performance AI model, Gemini Pro, for nearly seven months. The most advanced high-performance model currently available to users is Gemini Pro 3.1, which was released in mid-February. This contrasts with the aggressive moves from competitors, as OpenAI has already released GPT-6 and Anthropic has launched a series of high-performance models such as Claude Mythos, Fable, and Opus in succession.

At the annual developer conference “Google I/O 2026” held in May, Google had announced that it would launch Gemini 3.5 Pro in June. However, within the industry, it is widely believed that Google failed to develop the high-performance Gemini 3.5 Pro and missed the targeted release date. Instead, the company is expected to unveil the next-generation model Gemini 4, and to revise its strategy by releasing Gemini 4 Pro alongside it.

Manish Gupta, Senior Director of DeepMind at Google, is speaking about Google AI Gemini under the theme 'Google AI Opening New Possibilities' at 'Google for Korea' held on the 2nd at the Chosun Palace Hotel in Gangnam-gu, Seoul. Photo by Yonhap News

Rather than focusing on advanced models, Google has been responding to the market by releasing monthly updates to a lightweight version of its model—Gemini Flash—which targets general users. On the 2nd of this month (local time), the company unveiled its latest lightweight model, Gemini 3.8 Flash. While it came only three weeks after the previous Flash 3.7, consumers were disappointed to find that its performance fell short of the new “Muse Spark 1.3” model announced by Meta on the same day.

The latest version of Gemini lagged behind its competitors even in benchmarks that quantify model performance. Gemini 3.8 Flash received a score of 41 (for the high-end model) on the Artificial Analysis Intelligence Index (AAII), a global AI performance evaluation platform, failing to make the top 10. This is 12 points lower than top-tier competitor models such as Claude Fable (53 points, Anthropic) and GPT-6 Astra (53, OpenAI), and it also trails behind Chinese models like GLM-5.3 (45, Zhipu AI) and Kimi K3 (44, Moonshot AI). Even compared to other lightweight models, Gemini Flash scored lower than Muse Spark (48, Meta) and GLM-5.3-Flash (42).

Adding to these challenges, a wave of key AI researchers has left Google, heightening a sense of crisis within the company. According to Bloomberg and other major media outlets, four core AI talents recently departed from Google last month. Among them was Senior Scientist Jeff Dean, who joined Google in 1999 and had achieved legendary status in the industry—a move that sent shockwaves through the sector. Back in June, Gemini co-developer Noam Shazeer and Nobel Prize co-winner John Jumper also left, joining OpenAI and Anthropic, respectively.

While Gemini has struggled to deliver a new high-performance AI model, the company’s stock has also lost momentum. The share price of Alphabet, Google’s parent company, swelled to over $400 during intraday trading in May, but has since entered a downward trend and is currently hovering in the mid-$300 range.

This content was produced with the assistance of AI translation services.