AI Newsletter

AI Newsletter

Share this post

AI Newsletter
AI Newsletter
🤖 AI Agents Weekly: ChatGPT Agent, Gemini Embeddings, Agent Leaderboard v2, Voxtral, CRMAgent

🤖 AI Agents Weekly: ChatGPT Agent, Gemini Embeddings, Agent Leaderboard v2, Voxtral, CRMAgent

ChatGPT Agent, Gemini Embeddings, Agent Leaderboard v2, Voxtral, CRMAgent

Jul 19, 2025
∙ Paid
14

Share this post

AI Newsletter
AI Newsletter
🤖 AI Agents Weekly: ChatGPT Agent, Gemini Embeddings, Agent Leaderboard v2, Voxtral, CRMAgent
1
Share

In today’s issue:

  • OpenAI announces ChatGPT Agent

  • New agentic IDE called Kiro

  • Decart has launched MirageLSD

  • New survey on context engineering for LLMs

  • Windsurf releases Wave 11 (voice input and planning mode included)

  • Google introduces Gemini Embeddings

  • Enabling long-term, multimodal memory in LLM agents

  • How does increasing input tokens impact LLM performance

  • Agent Leaderboard v2

  • Top AI dev news, product updates, research papers, and more.



Top Stories

ChatGPT Agent

OpenAI introduces the ChatGPT Agent, a unified agentic system that integrates browsing, tool use, and reasoning to handle complex tasks end-to-end. It combines capabilities from the earlier Operator and Deep Research systems with new tools like a GUI-based web browser, terminal, and connectors (e.g., Gmail, GitHub). This allows it to autonomously conduct multi-step workflows, like planning events, analyzing competitors, or editing spreadsheets, on its own virtual computer, all while keeping the user in control via interruptibility and secure browser takeover.

  • Combines web interaction, code execution, and APIs into a single, coherent system that reasons and acts fluidly.

  • Achieves state-of-the-art results across benchmarks like WebArena (78.2% pass@1), BrowseComp (68.9%), and SpreadsheetBench (45.5% vs Copilot Excel’s 20.0%).

  • Outperforms humans in half of the cases on internal knowledge-work evaluations and excels in niche domains like investment banking modeling and green energy planning.

  • Emphasizes safety through prompt injection defenses, explicit user confirmation for real-world actions, and private browser sessions with deletion controls.

  • Treated as “High Bio/Chem capability” under OpenAI’s Preparedness Framework, with extra safeguards for misuse prevention and biosecurity.

ChatGPT Agent is rolling out to Pro, Plus, and Team users, with broader access coming soon.

Blog

Keep reading with a 7-day free trial

Subscribe to AI Newsletter to keep reading this post and get 7 days of free access to the full post archives.

Already a paid subscriber? Sign in
© 2025 elvis
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share