Last Week in AI #226 - Gemini 3, Claude Opus 4.5, Nano Banana Pro, LeJEPA
622 snips
Nov 30, 2025 
Michelle Lee, the brilliant Founder and CEO of Medra and a Stanford PhD, joins the discussion on the latest in AI. They dive into Google’s remarkable Gemini 3 and its groundbreaking coding IDE. The conversation swings to Anthropic's Opus 4.5, showcasing its impressive performance in coding tasks. Additionally, they explore innovative robotics with Sunday Robotics' Memo and a $600M funding round for Physical Intelligence. The episode also tackles AI's involvement in cyber espionage and new regulatory changes in Europe, highlighting the fast-evolving AI landscape.
AI Snips
Chapters
Transcript
Episode notes
Gemini 3 Is A Qualitative Leap
- Gemini 3 marks a qualitative leap, not just incremental improvement, across hard benchmarks like Humanity's Last Exam.
- Andrey Kurenkov notes Google trained Gemini 3 fully on TPUs and rolled it out broadly without major issues.
Image Editing Moves Beyond Pixels
- Nano Banana Pro (powered by Gemini 3) advances image editing toward accurate text and multi-image consistency.
- Michelle Lee and Andrey highlight its real-world uses like infographics and slide decks with imperceptible watermarking.
Opus 4.5 Raises The Bar
- Anthropic's Opus 4.5 outperforms Gemini 3 on several hard benchmarks and reduces cost significantly.
- It also shows stronger results on ARC/AGI-style intelligence puzzles, suggesting broader capability gains.