Google launches Gemini 3 with new coding app and record benchmark scores

On Tuesday, Google unveiled Gemini 3, its most sophisticated foundation model to date, now accessible via the Gemini app and AI search interface. This launch, occurring merely seven months after the introduction of Gemini 2.5, positions Gemini 3 as a formidable competitor among the top AI tools available today. The timing of this release is noteworthy, coinciding closely with OpenAI's rollout of GPT 5.1 and Anthropic's Sonnet 4.5 just two months prior, underscoring the rapid advancements in the development of frontier models. Google plans to introduce a more research-focused variant of this model, dubbed Gemini 3 Deepthink, to Google AI Ultra subscribers soon, pending further safety evaluations. "With Gemini 3, we observe a significant enhancement in reasoning capabilities," remarked Tulsee Doshi, Google's head of product for the Gemini model. He emphasized that the model exhibits a level of depth and nuance previously unseen. Gemini 3's reasoning prowess has already been validated through independent benchmarks, achieving a record score of 37.4 on the Humanity’s Last Exam benchmark, which assesses general reasoning and expertise. This score surpasses the previous high of 31.64 held by GPT-5 Pro. Furthermore, the model excelled in user satisfaction ratings on LMArena, leading the leaderboard. Currently, the Gemini app boasts over 650 million monthly active users, with 13 million software developers having integrated the model into their workflows. Additionally, Google introduced a Gemini-based coding interface named Google Antigravity, which enables multi-pane coding reminiscent of agentic IDEs like Warp or Cursor 2.0. This innovative tool merges a ChatGPT-style prompt window with a command-line interface and a browser window, allowing users to see the effects of the modifications made by the coding agent. "The agent can seamlessly interact with your editor, terminal, and browser to enhance your application-building experience," explained DeepMind CTO Koray Kavukcuoglu, highlighting the collaborative potential of this new tool.