Google's Gemini 3.5: Incremental or Disruptive?

The latest update to Google's flagship model is solid. But "solid" may not be enough anymore.

Published: 22 July 2026 Category: AI Models Sources: agenticaihype.substack.com

The Release

Google shipped Gemini 3.5 this week, and the initial reviews are... fine. Better reasoning than 3.0. Improved code generation. Longer context window — up to 2 million tokens, which is genuinely useful for certain enterprise applications. Better multimodal understanding, particularly for video. The benchmarks show improvement across the board, and the model is available through Google's usual channels: Vertex AI, the Gemini API, and increasingly tight integration with Workspace products.

"Fine" is not "exciting." And in 2026, "fine" may not be enough.

The Comparison

The problem is not that Gemini 3.5 is bad. It is that the competition is fierce. GPT-5.6 is the current benchmark leader. Claude Fable 5 is the enterprise favourite. Kimi K3 and Qwen 3.8 are the open-weight insurgents. Gemini 3.5 competes respectably with all of them on most tasks, but it leads on none. It is the definition of a solid third-place product in a market where first and second place get the attention.

Google's historical advantage — integration with Search, YouTube, Gmail, Maps — is still real but less decisive than it once was. Enterprise buyers increasingly choose models based on API quality, pricing, and safety guarantees, not pre-installed app integration. Consumers barely know what model powers their assistants. The brand advantage that Google enjoyed in search does not translate cleanly to AI model selection.

The Analysis

Google's AI strategy has been consistently competent and consistently overshadowed. DeepMind produces excellent research. Google Cloud offers excellent infrastructure. The Gemini models are well-engineered. But the company lacks the narrative momentum of OpenAI, the safety reputation of Anthropic, or the open-source credibility of Meta. It is the IBM of the AI era — technically capable, financially secure, and somehow always slightly behind the zeitgeist.

Gemini 3.5's multimodal capabilities are where Google still differentiates. The video understanding is genuinely better than competitors', reflecting years of YouTube infrastructure and research. For applications that require processing long videos — security footage analysis, content moderation, sports analytics — Gemini 3.5 is arguably the best choice. But these are niche applications compared to the text-and-code workloads that drive most AI revenue.

The Verdict

Gemini 3.5 is a good model that arrives at a time when "good" is the baseline. Google will sell plenty of API access. Enterprise customers already using Vertex will upgrade. The model will power Google Workspace features that millions use without knowing the model name.

But Google needs a breakthrough. It needs a Gemini 4.0 that redefines some category. It needs to own a dimension of AI that matters — not just participate in every dimension competently. Until then, Gemini will remain the safe choice, the default option, the model you use because it is there, not because you sought it out.

That is a profitable position. It is not a dominant one.

DEMYSTIFY