Gemini 3 Pro, Antigravity, Grok 4.1, Nano Banana Pro, SAM 3

Google took the week. After months of rumor, Gemini 3 arrived on Tuesday.

Gemini 3

Google released Gemini 3 Pro as its most capable model and shipped it into Search, the Gemini app, AI Studio and Vertex on day one. That last part is the shift. Earlier Gemini launches reached the API first and the products later. Google can ship a flagship into a search engine used by billions on the same day, and none of the other labs have that distribution.

The company reports top scores on several reasoning and multimodal benchmarks. I'd wait for independent comparisons on real tasks before concluding anything, since every lab leads on its own chosen charts. The part that made me stop was the focus on agentic work and long-horizon tool use.

Antigravity

The sibling launch was Google Antigravity, an agent-first development environment. Agents get access to the editor, the terminal and a browser, and they produce what Google calls artifacts, such as task lists, plans, screenshots and browser recordings, so you can verify what happened. I've written before about the trust problem with delegation, and I think this is the right instinct. You can't read every step an agent took. You can look at an artifact that shows the result.

The rest of a loaded week

xAI released Grok 4.1, aimed at tone, creative writing and fewer hallucinations. OpenAI followed with GPT-5.1-Codex-Max, a coding model that works across multiple context windows on long tasks. I link the Codex repository as the primary artifact because OpenAI's own launch page blocked my link check.

Google also released Nano Banana Pro for image generation and editing on top of Gemini 3, with better text in images. Meta shipped Segment Anything Model 3, which segments objects from a text description. The Allen Institute released Olmo 3, a fully open family with data, code and checkpoints.

Anthropic announced partnerships with Microsoft and NVIDIA: it commits to Azure compute while the two companies plan to invest in Anthropic. The labs now share chips, clouds and investors so tightly that picking a side is hard.

What I'd watch next

Anthropic's reply. Gemini 3 raised the bar on coding benchmarks, and Claude's next model has to answer.