Gemini 3.5

Gemini 3.5 is a family of artificial intelligence models developed by Google DeepMind, introduced at Google I/O in May 2026 under the description "frontier intelligence with action". The series opened with Gemini 3.5 Flash, positioned at flagship-level capability but Flash-tier speed for agentic and coding workloads, with a larger 3.5 Pro announced to follow.^[1]

History

Google published the Gemini 3.5 announcement on 19 May 2026 in a post authored by CTO Koray Kavukcuoglu, Chief Scientist Jeff Dean, and Vice Presidents Oriol Vinyals and Noam Shazeer. The company called the series "a major leap forward in building more capable, intelligent agents" and released 3.5 Flash the same day to consumers, developers and enterprises.^[1]

Flash was made available globally through the Gemini app and AI Mode in Google Search, to developers through the Google Antigravity agent development platform and the Gemini API in Google AI Studio and Android Studio, and to businesses via the Gemini Enterprise Agent Platform. Google said 3.5 Pro was already in internal use, with rollout planned for the following month.^[1] In October 2026 Google made Gemini 3.5 Flash the default model of Search's AI Mode worldwide, citing its frontier performance for agents and coding; the Flash line has since advanced further within the 3.5 generation and beyond.^[3][4]

Key concepts and technology

Gemini 3.5 models are natively multimodal reasoning models built on the transformer architecture, trained across text, image, audio and video. Headline results Google reported for 3.5 Flash included 76.2% on Terminal-Bench 2.1 (agentic terminal task completion), 1,656 Elo on GDPval-AA (real-world occupational work), 83.6% on MCP Atlas (tool use through the Model Context Protocol) and 84.2% on CharXiv Reasoning for multimodal understanding — outperforming Google's earlier flagship Gemini 3.1 Pro while producing output tokens about four times faster per second than other frontier models.^[1] The model card describes 3.5 Flash as the next iteration in the Gemini 3 series of highly capable, natively multimodal, reasoning models.^[2]

The Flash versus Pro tiering follows Google's established pattern: Flash models prioritise latency and cost for high-volume and interactive use, while Pro models spend more computation on difficult reasoning. The generation marks a stated pivot from pure benchmark competition toward long-horizon agentic task performance — planning and executing multi-step work with less human supervision.

Applications and impact

Gemini 3.5 underpins Google's consumer AI surfaces — Search AI Mode, the Gemini app and Android features — as well as enterprise agent platforms and third-party applications built on the Gemini API. Its agentic emphasis reflects an industry-wide shift from chatbots to software that browses, codes and researches semi-autonomously, and it sets the default quality bar for the world's most-used search engine.

>Key Takeaways

• Gemini 3.5 launched at Google I/O in May 2026, starting with 3.5 Flash and with 3.5 Pro announced to follow.
• It is tuned for agentic coding and long-horizon tasks, and became the default model for Google Search's AI Mode.
• Flash-tier speed with near-flagship scores made it Google's volume model for consumers, developers and enterprises.

See Also

• Gemini
• Google DeepMind
• Reasoning Models
• Model Context Protocol
• Large Language Models