In partnership with

Welcome back!

Google just pulled ahead again. With the release of Gemini 2.5 Pro 06-05, the tech giant now commands the top spots on nearly every AI leaderboard. From coding and logic to long-form writing and creative tasks, it looks like Google’s racing itself now. Let’s unpack…

Today’s Summary:

• 🚀 Google’s Gemini 2.5 Pro 06-05

• 🔧 ChatGPT adds meeting recording tools

• 🧲 DeepMind physics prompting

• ⚖️ Court forces ChatGPT to retain chats

• 🔬 AI breast cancer tool gets FDA nod

• 🤖 Reddit sues Anthropic over scraping

• 🛠️ 2 new tools

TOP STORY

Gemini 2.5 Pro update widens Google’s AI lead

The Summary: Google released Gemini 2.5 Pro version “06-05”, in preview ahead of general rollout. With a +24 Elo jump on LM Arena and +35 on WebDev Arena over its May version, it now holds the #1 spot across nearly all major AI benchmarks. The update boosts coding, reasoning, and creative performance, with broad improvements in style and formatting.

Key details:

• Gained +24 Elo points over May version, now #1 on LM Arena at 1470

• Jumped +35 on WebDevArena to 1443, beating Claude 4 Opus and all other code-first models, except on the SWE-bench Verified (Agentic coding), where Claude still leads

• #1 across every LM Arena category: Hard, Math, Coding, Creative, Vision, Long-form, Multi-turn

• Available now on Gemini app, Vertex AI, and AI Studio

Why it matters: Google now holds both the #1 and #2 spots on LM Arena with Gemini 2.5 Pro edging out its own earlier version, no other lab is close. Gemini is tightening its grip on core reasoning, raw intelligence and usability. It looks like Google isn’t racing OpenAI anymore, but it’s racing itself.

Try it in AI Studio

FROM OUR PARTNERS

Work Smarter with These Prompts

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide

OPENAI

ChatGPT adds meeting recording and more integrations

The Summary: OpenAI released new business features for ChatGPT, including Record Mode for meetings, real-time transcription, and deep research across connected apps like Google Drive, Outlook, and Dropbox. Available across most paid plans, these tools let ChatGPT pull directly from company files and conversations. Custom connectors are also available.

Key details:

• Record Mode auto-transcribes meetings and suggests actions, rolling out initially on macOS for Team users, with more plans to follow soon

• Audio files from recordings are auto-deleted after transcription

• ChatGPT now supports connectors to Google Drive, Outlook, Dropbox, and more, both in regular chats and in Deep Research (not available in the EU, CH and UK due to regulations)

• Synced connectors, starting with Google Drive, preload company files for faster answers; available to Enterprise and Edu users

• Custom connectors can be built using Model Context Protocol (MCP), allowing Pro, Team, and Enterprise users to query internal data or third-party systems

Why it matters: OpenAI is continuing to push ChatGPT from sidekick to a central operating layer for work. By embedding it into files, systems, and meetings, it becomes the tool that “knows everything you said and wrote”. The record and transcribe feature handles meeting documentation. MCP opens the door to an app store-like ecosystem for custom enterprise AI.

FROM OUR PARTNERS

Unbiased Business Insights, Every Week

Get a weekly knowledge digest of business insights from 1440’s Business & Finance newsletter. Expect concise rundowns, context-rich visuals, and curated links to keep you ahead of the finance curve. Join 1440 for crisp explanations of business trends —no MBA required.

Sign Up Here

GOOGLE

DeepMind’s Force Prompts teach video models physics

The Summary: Researchers at DeepMind and Brown University found a way to train video generation models to simulate real-world physical reactions. Their “force prompting” method lets users control motion by applying direction and magnitude vectors to static images. The technique shows surprising generalization across materials, settings, and motion types.

Key details:

• The model uses vector fields to simulate global (wind) and local (poke) physical forces in AI-generated videos

• Learned object mass: heavier items move less under the same force, like a full laundry basket lagging behind an empty one

• Despite training only on simple scenarios, it generalized to fog, confetti, paper lanterns, and fluid-like motion

Why it matters: This research shows how AI might start learning physics by seeing. Instead of relying on scripted motion, it taps into visual priors built into the model. This could lead to more interactive media tools, where creators guide movement with simple inputs like push and pull.

Research paper

QUICK NEWS

Quick news

• Court forces ChatGPT to save deleted chats

• Clarity gets FDA approval for AI breast cancer risk prediction

• Reddit sues Anthropic for scraping site content to train Claude

TOOLS

🥇 New tools

• Wispr Flow - Effortless voice dictation in every application

• Mistral Code - Enterprise AI coding with full control & security