Google announces the 'Gemini 3.5' series, starting with the lightweight 'Gemini 3.5 Flash'.



Google held its annual developer conference, '

Google I/O 2026, ' starting on May 19, 2026, and announced its AI series, ' Gemini 3.5, ' during the keynote address. Google described Gemini 3.5 as 'a new family of models that combines cutting-edge intelligence and autonomous behavior,' and announced the availability of Gemini 3.5 Flash as the first product in the series.

Gemini 3.5: frontier intelligence with action
https://blog.google/intl/ja-jp/company-news/technology/gemini-3-5/

Gemini 3.5 Flash might be fast enough for gen AI to make sense - Ars Technica
https://arstechnica.com/google/2026/05/google-announces-agent-optimized-gemini-3-5-flash-and-a-do-anything-model-called-omni/

Gemini 3.5 Flash is available through the Gemini app and Google Search's 'AI mode.' It can also be accessed from the agent development platform Google Antigravity, as well as the Gemini API in Google AI Studio and Android Studio, the Gemini Enterprise Agent Platform, and Gemini Enterprise.

The Gemini 3.5 Flash maintains the overwhelming speed that is characteristic of the Flash series, while offering advanced intelligence comparable to flagship models in multiple dimensions. Google states that it is 'the most powerful model to date in the field of building and coding AI agents.'

In fact, the Gemini 3.5 Flash achieved higher scores than the Gemini 3.1 Pro in Terminal-Bench 2.1 , a benchmark for complex coding and agent evaluation, GDPval-AA , a rating that measures the inference ability of LLMs, and MCP Atlas , which evaluates the collaborative ability of multi-agent systems. Furthermore, it achieved an excellent score of 84.2% in CharXiv Reasoning , a benchmark that measures multimodal understanding including academic charts and graphs. Its output tokens per second are four times faster than other state-of-the-art models.



According to

Artificial Analysis , a third-party organization that independently evaluates the performance and cost of AI models, Gemini 3.5 Flash has been positioned in the top segment, the Leader category. Google is touting this as proof that it is no longer necessary to sacrifice model quality or accuracy to achieve superior processing speed.



Gemini 3.5 Flash, when combined with the Google Antigravity harness, evolves into a powerful engine capable of coordinating and operating multiple sub-agents. This will enable it to handle extremely demanding, large-scale use cases, allowing for the reliable and smooth execution of complex multi-step workflows and coding tasks while maintaining cutting-edge performance under human supervision.



Furthermore, the multimodal capabilities will enable the creation of more expressive and interactive web UIs and graphics.



Google announced Gemini Spark, a personal AI agent, at Google I/O 2026. Gemini Spark fully utilizes Gemini 3.5 Flash. At the time of writing, Gemini Spark is being rolled out to trusted testers, and a beta version is planned to be available to Google AI Ultra users in the United States by June 2026. Furthermore, Gemini 3.5 Flash is also being applied to Google's revamped AI search.

Gemini 3.5 was developed in accordance with Google's Frontier Safety Framework , the most advanced safety standard set by Google. By significantly strengthening defenses and safety measures regarding cybersecurity and CBRN (chemical, biological, radioactive, and nuclear) materials, the risk of generating harmful content has been reduced to the absolute minimum. At the same time, the probability of incorrectly refusing to answer safe questions has also been greatly reduced. Google emphasizes that this was achieved by introducing advanced safety training and mitigation measures, such as Interpretability tools that verify and understand the internal reasoning process before the AI outputs an answer.

Google has also revealed that it is developing Gemini 3.5 Pro. It appears that Google has already begun using Gemini 3.5 Pro internally, and it is scheduled to be released to the public in June 2026.

in AI, Posted by log1i_yk