Google skips to Gemini 3.5, signaling major architectural overhaul

Google is developing Gemini 3.5, according to a DeepMind employee who confirmed the version number this week. The jump from 2.0 directly to 3.5 signals a significant architectural overhaul rather than an incremental release—comparable to the 1.0-to-2.0 transition that brought native image and video understanding, longer context windows, and stronger coding performance.

Gemini 2.0 shipped in December 2024. Skipping intermediate releases (2.5 or 3.0) typically indicates performance gains substantial enough to justify the version leap. Google has not disclosed technical specifications, parameter counts, training data, or capability details.

The confirmation puts Google in direct competition with OpenAI's GPT-5 development cycle and Anthropic's Claude Opus series, both expected to ship major updates in 2025. Google has historically unveiled major Gemini versions at its May I/O conference, though the company has launched models outside that window when ready.

Gemini models remain closed-weight and API-only, limiting their use to practitioners building on Google's cloud infrastructure. The company has not signaled any shift toward open-weight releases, even as Meta continues shipping Llama checkpoints and Chinese labs release unrestricted multimodal models. For developers on Google's platform, the 3.5 confirmation suggests the current 2.0 API will remain stable through at least mid-2025, with migration guidance likely announced closer to launch.