Google AI

Google's developer slate: Antigravity 2.0 desktop plus a CLI and SDK for programmatic control of its coding agent, managed agents hosted in the Gemini API, Kotlin Android builds publishable from AI Studio, WebMCP in a Chrome 149 origin trial, and Chrome DevTools agents can drive autonomously.

Gemini Spark is a persistent personal agent that takes action across Gmail, Docs, and Slides and keeps running after you close your laptop. Daily Brief, a packaged agent producing a goal-based digest with suggested next steps, rolls out first to US subscribers, with Spark following next week.

A live model built for realtime interaction: response latency close to natural dialogue, better task completion in noisy environments, and improved complex instruction following. Available via the Gemini Live API and AI Studio in preview, plus Gemini Live and Search Live.

Google's weekly recap covers EmbeddingGemma 2, a natively multimodal open embedding model for on-device use, Nano Banana 2.1 spot editing, a Gemini cloud agent for business, and a global AI-content detector portal.

Google reports 77.1% on ARC-AGI-2 for Gemini 3.1 Pro, more than double the 3 Pro result, on a benchmark built around unfamiliar logic patterns. The model is shipping across Google's consumer and developer products.

A 3.5 Flash variant tuned for finding and fixing vulnerabilities at scale, deployed inside Google's CodeMender agent and measured on benchmarks like CyberGym. Access stays limited to governments and trusted partners through a pilot rather than the open API.

Google's new image model draws on Gemini's world knowledge plus web search to render specific landmarks and objects accurately, localizes assets by region, edits aspect ratios, upscales to 2K and 4K, and keeps multiple images coherent within one concept. Available day one in AI Studio, Vertex, Antig

Nano Banana 2 Lite returns text-to-image output in under four seconds as Google's cheapest image model, while Gemini Omni Flash enters public preview for cost-efficient video generation and conversational editing. Both are callable from the Gemini API and AI Studio.

Project Genie can now seed interactive world generation from a Google Maps Street View location, using that imagery to fill in what sits behind the camera. Worlds are produced frame by frame from a world description plus the user's actions, in a prototype available to Ultra subscribers.

Google's new workhorse model is priced at $0.75 per million input tokens and $3.75 per million output through year end, half the 3.6 Flash rate. It emphasizes multi-step planning and tool calls, and is available in the Gemini API, Antigravity, and Gemini Enterprise.

Google's weekly recap: Gemini 3.8 Live and Extended Thinking audio models ship, Dreambeans personalized story feed reaches GA, CC expands into a shared household agent, Google Pics image co-creation goes GA, and DeepMind's AlphaGenome Atlas launches for genomics.

Gemini 3 Flash can now treat image understanding as an agentic loop, generating and executing Python to manipulate an image mid-reasoning, appending the transformed result to its context window, then answering. Google reports a consistent 5 to 10 percent gain across most vision benchmarks.

Google Labs expands Project CC into a family group agent. It gets its own verified Google account, receives only forwarded mail and shared files, and runs Gemini on isolated cloud machines. Access is by waitlist.

MedGemma 1.5 4B improves on 3D imaging and reasoning over medical records, while TranslateGemma ships multimodal open translation models at 4B, 12B and 27B across 55 language pairs. Both are on Hugging Face, with MedGemma on Vertex AI and TranslateGemma on Kaggle.

Google opened a public SynthID portal for checking whether images, video or audio carry its watermark, covering Google models and partners including OpenAI, NVIDIA and Kakao, with Apple to follow. Google cites 180 billion watermarked images and videos and about 1 million daily verifications.

Google's cyclone model predicts track, intensity and size together instead of splitting global physics models from zoomed-in regional ones, buying roughly a full extra day of warning and generating up to 1,000 simulations per storm. Published in Nature, used operationally by the US National Hurrican

Chrome's Gemini side panel adds multi-step auto browse for tasks like forms and appointments, connections to Gmail, Calendar, YouTube, Shopping and Flights, inline image generation, and support for Google's Universal Commerce Protocol so agents can execute transactions. US desktop first, in preview

Google's weekly recap covers new Gemini TTS models, Gemini 3.8 Live's real-time avatar feature, NotebookLM's Interactive Learning Overviews and mobile Live Chat, and progress on Project Suncatcher, a prototype satellite testing TPUs in orbit for solar-powered compute.

Google's new Nano Banana-based image editor lets Pro/Ultra subscribers isolate and edit objects, translate in-image text, and generate variants from one prompt, with Docs and Slides integration already live and Drive support coming soon.

Living Models paired Gemma 4, acting as a coordinator that writes code and filters data, with the BOTANIC-1 plant DNA model. It ranked a melon yield mutation first of 2,494 candidates in under four minutes, a task that normally takes breeding trials over years.

Google opens three experimental research tools: hypothesis generation with Co-Scientist, computational discovery with AlphaEvolve and ERA, and literature synthesis with NotebookLM. Science Skills wires 30+ life science models and databases into agent platforms like Antigravity.

Google released Gemini 3.8 Flash TTS and Flash-Lite TTS: expressive voice models supporting 100+ languages, 2,000+ preset voices, line-by-line delivery control, and inline cues like <laughs>. Flash targets studio production, Flash-Lite targets real-time voice agents.

Google's new speech-to-text model covers 85+ languages, strips filler words, formats unstructured speech, and can use screen context to execute voice commands. It ships in the Gemini app on macOS, Gboard on Android, and the Gemini API via AI Studio and Antigravity.