Quietly Drifting Away from Gemini
I’ve used Google’s Gemini models for some time, but lately I’ve found myself increasingly underwhelmed by the quality of their output. Before I knew it, I was reaching for competing AI models in almost every scenario, and Gemini had quietly slipped out of my routine. The gap between Gemini and rival services seems to be widening by the day.
In this article, I summarize what’s happening with Gemini, and what it tells us about the state of the AI industry as of July 2026.
Gemini 3.5 Pro Keeps Slipping by Months
The single biggest factor is the major delay affecting Google’s next flagship model, “Gemini 3.5 Pro.”
At Google I/O in May 2026, Google announced that Gemini 3.5 Pro would be generally available by June. Yet here we are in late July, and the official release still hasn’t materialized. According to reporting by Bloomberg, the model’s coding performance is falling short of internal targets, and development is running months behind the original plan. The news even triggered a dip in Alphabet’s share price.
The only model currently in general availability is the lightweight “Gemini 3.5 Flash”—not the flagship. In other words, the “latest Gemini” most of us encounter day to day is not the top-tier model Google originally intended to ship, but the step below it. That may explain the lingering sense of “good, but not quite there.”
Meanwhile, the AI Battleground Is Shifting to the Enterprise
Beyond the Gemini-specific issues, the structural shape of the industry is changing. In recent years, the center of gravity for AI revenue has been migrating from consumer-facing chat services to enterprise development workflows.
In the enterprise LLM API market, Anthropic is gaining real traction. Its market share reportedly approaches 40%, and within coding-specific spending it’s said to capture roughly 40–50%, while both OpenAI and Google hover around 20%.
Meanwhile, OpenAI maintains overwhelming scale on the consumer side via ChatGPT (with over 900 million weekly active users) and is investing aggressively in the enterprise market as well. Still, at least for now, Anthropic stands out most in the enterprise LLM API and coding segments. Rather than “OpenAI and Anthropic as a two-horse race,” it’s more accurate to say: OpenAI holds overwhelming scale on the consumer side, while Anthropic leads on the enterprise and coding front.
Chinese Models Are Rapidly Expanding Share
Another trend impossible to ignore is the rise of China-origin models.
On the developer platform OpenRouter, U.S. usage of Chinese-made AI models has consistently exceeded 30% week-over-week since the start of 2026, with some reports putting it close to 50% at peak. Compared with an average of under 10% in the first half of 2025, this is a remarkably rapid shift.
- Kimi K3 (Moonshot): Demand surged immediately after release, prompting the company to temporarily pause new registrations citing compute constraints.
- Qwen 3.8 (Alibaba): Rapidly closing the performance gap with leading Western models.
- GLM-5.2 (Zhipu AI): Recorded the fastest adoption growth tracked on Vercel’s developer platform in 2026.
- DeepSeek: AI startup Lindy migrated entirely from Anthropic’s models to DeepSeek, emphasizing the cost savings.
Armed with low prices and strong performance, these models are becoming the safety valve for Western companies squeezed by rising per-token costs.
Conclusion: Gemini’s Stagnation Isn’t “Just a Feeling”
To summarize:
- Gemini’s flagship model has been delayed for months on the grounds of missing internal targets, leaving only the lightweight version available.
- The AI industry’s main battleground is shifting to the enterprise, where Anthropic has the edge, while OpenAI defends overwhelming scale on the consumer side but is reportedly struggling in the enterprise arena.
- China-origin open-weight models are rapidly gaining share on both price and performance.
In other words, the feeling of “I’ve stopped using Gemini lately” was backed by concrete facts—specifically, the lag in Google’s own model development. As the AI industry as a whole rapidly becomes more multipolar, how Google recovers from this stagnation is an open question. The official release of Gemini 3.5 Pro looks likely to be the first real test.