Gemini 3.5 is a generation of proprietary large language models developed by Google DeepMind, announced on May 19, 2026 as the successor to Gemini 3 within the Gemini family.1 The generation launched in an unusual order: the lighter-weight Gemini 3.5 Flash shipped on announcement day, while the flagship Gemini 3.5 Pro was described as in internal use with a public rollout planned "next month." As of July 2026 the Pro model had not been released.2
Gemini 3.5 Flash
Gemini 3.5 Flash was released at Google I/O 2026 and immediately became the default model in the Gemini app and in AI Mode in Google Search. Google positioned it as delivering intelligence that "rivals large flagship models" while retaining Flash-tier speed and cost, with particular emphasis on agentic workflows and coding. Developer-reported figures included 76.2% on Terminal-Bench 2.1, 83.6% on MCP Atlas, 84.2% on CharXiv Reasoning, an Elo of 1656 on GDPval-AA, and output speeds described as roughly four times faster than competing frontier models.1 The launch highlighted long-horizon agent execution through the Google Antigravity platform and introduced Gemini Spark, a continuously running personal agent built on 3.5 Flash. As with earlier generations such as Gemini 2.5 and Gemini 1.5, Google disclosed neither parameter counts nor architectural specifics, and the model is natively multimodal.
Gemini 3.5 Pro
Gemini 3.5 Pro missed a widely reported July 17, 2026 launch target and remained in limited Vertex AI preview, with press accounts (citing Bloomberg) attributing the delay to efforts to improve coding performance; an August window was rumored as of mid-July 2026.23 Media reports described an expected 2-million-token context window and a built-in "Deep Think" mode for extended deliberate reasoning, extending Google's use of test-time compute; Google had not confirmed these specifications. Press coverage framed the staggered rollout against contemporaneous frontier releases such as GPT-5.1 and Claude Opus 4.6.
Availability
Gemini 3.5 Flash is proprietary and accessible through the Gemini app, AI Mode in Search, the Gemini API via Google AI Studio and Android Studio, Google Antigravity, and Gemini Enterprise. Weights are not distributed, consistent with every prior Gemini release since Gemini 1.0.