🧠 Gemini Gets Physical, Codex Levels Up, and Open Models Race Ahead

🔥🔥🔥 Three Curated AI Stories

1. Gemini 3 Deep Think turns sketches into printable objects

Google released a major upgrade to its specialized reasoning mode, Gemini 3 Deep Think, designed for multi-step problem solving and complex analytical tasks. Deep Think is built to help researchers interpret dense information, and engineers model physical systems through code, not just generate quick answers.

The headline demo shows what that reasoning looks like in practice: you can sketch an object, and Deep Think analyzes the geometry, constructs the full 3D model, and generates a printable file. It’s less about the sketch itself and more about the system’s ability to reason through spatial and structural constraints to produce something usable.

For engineers and product teams, this shortens the path from idea to prototype. For researchers and educators, it opens new ways to explore complex systems interactively. And for makers, it turns rough concepts into testable objects without heavy CAD work.

This is rolling out now to Google AI Ultra subscribers. Select “Deep Think” in the tools menu to start experimenting.

Implementation details and examples are here.

2. OpenAI introduces GPT-5.3 Codex Spark

OpenAI announced GPT-5.3 Codex Spark, a new coding-focused model designed to assist with writing, understanding, and iterating on software. It’s positioned as a stronger collaborator for complex development workflows.

For developers, this reduces friction in debugging, refactoring, and rapid prototyping. For startup teams, it speeds up building internal tools and MVPs without expanding headcount. And for technical learners, it acts as a guided coding partner that explains decisions instead of just generating output.

If you’re integrating AI into your dev stack, OpenAI’s release notes include examples and setup guidance.

3. MiniMax launches M2.5 for real-world productivity

MiniMax released M2.5, an open-source frontier model aimed at practical workplace tasks, with strong performance in coding, search, agent tool-calling, and office workflows. It also runs faster on complex jobs and is priced to make long-running agents more economically viable.

For teams building AI agents, this lowers experimentation costs and enables longer autonomous workflows. For business operators, it opens the door to scalable automation in research, reporting, and repetitive office tasks. And for developers, it provides an open model you can customize without relying entirely on closed platforms.

MiniMax provides direct access to its agent environment and API here:
—> MiniMax Agent 
—> API
—> CodingPlan

I used the MiniMax Agent and tested it by asking it to conduct a comprehensive analysis of Starbucks’ financials:

If you want the prompt I used or the results, reply to this newsletter, and I will send them your way!

🔥🔥 Two Tools Worth Trying

1. 🧩 Manus Skills for Projects

Manus added Skills to Projects, letting teams build reusable workflows inside a shared workspace. You can standardize how tasks get done and onboard new teammates faster by packaging repeatable processes as skills.

It’s especially useful for teams that run the same sequences over and over, from research pipelines to client onboarding.

2. ✨ Gemini tricks inside Google Workspace

Gemini now hides a few powerful shortcuts inside everyday Google tools. You can summarize an entire Google Doc by typing @Gemini, summarize folders directly in Drive, combine multiple Drive files as context for emails or actions, and update slides using information from a YouTube video.

These features turn Gemini into a lightweight assistant embedded in your daily workflow, not a separate app you have to switch to.
 

🔥🔥 One Trending AI Mega Prompt

Give your OpenClaw agent a real personality

This workflow uses a viral prompt to rewrite your OpenClaw (or ClawdBot) system file so your AI agent develops a distinct voice and behavior style. Instead of sounding generic, it responds with a consistent tone, opinions, and personality.

First, here’s the prompt:

Here’s how to implement it:

1. Open your OpenClaw or ClawdBot environment and locate the system or SOUL file.

2. Apply the personality prompt from the shared image to rewrite that file.

3. Save the updated configuration and restart your agent.

4. Interact with it and refine the tone based on your preferences.