Ur_Kowsar

#GoogleLaunchesGemini3.5Flash At Google I/O 2026, Google DeepMind introduced Gemini 3.5 Flash, the first model in the new Gemini 3.5 family. This is an unusual announcement: the Flash model, traditionally focused on speed and low cost, for the first time in the history of the line surpassed the flagship Gemini 3.1 Pro in key indicators of coding and agent tasks. The model is available today in the Gemini app, AI Mode in Google Search, Google AI Studio, Antigravity 2.0 and Gemini API. The key indicator is speed: according to Google CEO Sundar Pichai, Gemini 3.5 Flash issues 289 tokens per second, which is four times faster than competing frontier models. In benchmarks, the model scored 76.2% on Terminal-Bench 2.1 (coding), 1656 Elo on GDPval-AA (real agent tasks) and 83.6% on MCP Atlas (scalable tooling). For comparison: Gemini 3.1 Pro scored 1317 on GDPval-AA — Gemini 3.5 Flash outperforms it by 26%. At the same time, the model is inferior to 3.1 Pro in complex academic reasoning and dense tasks with long context — so for the most complex one-time requests 3.1 Pro remains relevant. Gemini 3.5 Flash is built for agent-based and long-running tasks: it can plan and execute work by parallel subagents, support complex multi-step workflows for a long time, and build more complex interactive web interfaces. It is on it that Gemini Spark and Antigravity 2.0 — two key agent tools, also announced at I/O 2026 — work. The price for developers is $1.50 per million incoming and $9.00 per million outgoing tokens, which is 40% cheaper than Gemini 3.1 Pro with higher performance in agent scenarios. Context window — 1 million tokens. $BNB