Hello from The AI Night,
Today in AI:
Google's New Flagship Model, Gemini 3.1 Pro is here
Google Labs Launches Free AI Product Photography Tool “Photoshoot”
Anthropic Launches Claude Code Security to Find Vulnerabilities in Codebases
Image Source: Google blog
Here's the deal: Google released Gemini 3.1 Pro, an upgraded reasoning model now rolling out across the Gemini app, API (AI Studio), Vertex AI, Gemini CLI, NotebookLM, Google Antigravity and Android Studio. It serves as the core intelligence behind last week's Gemini 3 Deep Think update.
The Breakdown:
Scored 77.1% on ARC-AGI-2 (abstract reasoning puzzles) more than 2x the 31.1% from Gemini 3 Pro and ahead of Sonnet 4.6 (58.3%), Opus 4.6 (68.8%) and GPT-5.2 (52.9%)
Leads or matches top models on 15+ benchmarks including GPQA Diamond (94.3%), SWE-Bench Verified (80.6%), Terminal-Bench 2.0 (68.5%) and MCP Atlas (69.2%)
Currently in preview for developers. Consumer access requires Google AI Pro or Ultra plans
Google signals further improvements in agentic workflows before general availability
The bigger picture: Gemini 3.1 Pro positions Google at or near the top of most major reasoning and coding benchmarks in a single model. For developers building agentic applications, it widens the set of viable foundation models with strong tool use and long context performance.
Image Source: Anthropic blog
Here's the deal: Anthropic released Claude Code Security, a new capability built into Claude Code on the web, now in limited research preview. It scans codebases for security vulnerabilities and suggests patches for human review, targeting complex bugs that traditional static analysis tools typically miss.
The Breakdown:
Instead of matching known patterns, it reads and reasons about code like a human security researcher, tracing data flow and understanding component interactions
Every finding goes through multi stage verification where Claude attempts to disprove its own results before surfacing them
Findings include severity ratings and confidence scores, nothing is applied without human approval
Using Claude Opus 4.6, Anthropic's team found over 500 vulnerabilities in production open source codebases some undetected for decades
Available now to Enterprise and Team customers with free expedited access for open source maintainers
The bigger picture: This shifts AI-powered vulnerability detection from research demos to a production tool defenders can actually use. As AI lowers the barrier for attackers to find exploitable weaknesses, tools like this could help security teams close gaps before they're exploited.
Image Source: Google blog
Here's the deal: Google Labs introduced Photoshoot, a new feature inside Pomelli, its free marketing tool for small and medium sized businesses. Photoshoot uses what Google calls "Nano Banana" image generation and stored business context ("Business DNA") to turn basic product photos into studio quality marketing images.
The Breakdown:
Upload any product photo, choose a template (studio, lifestyle or AI suggested), and Photoshoot generates on brand professional images automatically
Includes editing capabilities like background swaps and style transfer between reference images
New campaign tools let users upload images or enter a product URL to generate promotions grounded in real product data
All outputs are downloadable or stored in Business DNA for reuse across campaigns
Available free in the US, Canada, Australia and New Zealand
The bigger picture: This directly targets the gap where small businesses lack budget for professional product photography and design. If the output quality holds up, it removes a real friction point between having a product and marketing it effectively especially for e-commerce and social
What else you need to know:
Google Chrome added a split view for side by side tabs, built in PDF annotation with highlights and notes, and a "Save to Google Drive" option that sends downloads directly to Drive without re uploading.
OpenAI upgraded Code Blocks in ChatGPT to support in chat code editing, split screen review, full screen editing and live preview of diagrams, Mermaid charts and mini apps.
Cursor shipped agent sandboxing across macOS, Linux and Windows that lets coding agents run freely inside controlled environments, reducing approval interruptions by 40% compared to unsandboxed agents.
Cursor launched a plugin marketplace with partners like Figma, AWS, Stripe and Linear, letting developers install packaged skills, subagents and MCP servers in one click.
Replit launched Animation, a tool powered by Google Gemini 3.1 Pro that generates professional React based animated videos from text prompts with exports up to 1080p at 60fps for around $1.
That’s it for today’s edition of The AI Night.
Our goal is to cut through the noise, surface what actually changed, and explain why it matters.
If this was useful, you’ll get the same signal here tomorrow.