🚀 OpenAI Launches ChatGPT Pulse

PLUS: GDPVal Tests on Real-Work Tasks

Author

The Summary AI
September 29, 2025

In partnership with

Welcome back!

OpenAI just launched ChatGPT Pulse, an experiment where the chatbot doesn’t just answer but proactively suggests workouts, travel plans, or project next steps. It’s like a personal feed curated by AI and delivered daily. Let’s unpack…

Today’s Summary:

  • 🔥 OpenAI debuts proactive ChatGPT Pulse

  • ⚡️ Google upgrades Gemini 2.5 Flash

  • đź“Š New GDPval benchmark tests real-world tasks

  • đź’ˇ Google Labs unveils Mixboard

  • 🎥 Alibaba Wan2.5 syncs video & audio

  • 🦾 Google launches Gemini Robotics 1.5

  • 🛠️ 2 new tools

TOP STORY

OpenAI introduces ChatGPT Pulse for proactive personalization

The Summary: OpenAI has launched ChatGPT Pulse, a feature that flips the chatbot from reactive to proactive. Instead of waiting for prompts, Pulse learns from past chats, feedback, and app connectors, then delivers a daily set of personalized cards. Pulse is an experiment with AI that initiates conversations rather than waiting to be asked.

Key details:

  • Pulse surfaces daily cards on topics mentioned in recent chats, ranging from workouts and recipes to trip and project planning

  • OpenAI positions Pulse as an evolution of its earlier “Tasks” tool

  • Gmail and Google Calendar can be linked (disabled by default)

  • Cards will auto-delete unless saved

  • Currently limited to Pro subscribers on mobile, coming later to Plus

Why it matters: Pulse looks like the prototype of an AI version of a social feed. Fidji Simo, who leads OpenAI’s applications and once headed Meta’s social feed products, is experimenting on a proven pattern, with the difference that Pulse is framed around utility and productivity. OpenAI is testing whether people see value in ChatGPT delivering content proactively and unprompted.

FROM OUR PARTNERS

The Smartest Way to Turn Speech Into Writing

Smart dictation that understands you

Typeless turns your raw, unfiltered voice into beautifully polished writing - in real time.

It works like magic, feels like cheating, and allows your thoughts to flow more freely than ever before.

With Typeless, you become more creative. More inspired. And more in-tune with your own ideas.

Your voice is your strength. Typeless turns it into a superpower.

GOOGLE

Google releases improved Gemini 2.5 Flash

The Summary: Google has rolled out new versions of Gemini 2.5 Flash and Flash Lite, now available in AI Studio and Vertex AI API. The updates bring faster response times, sharper reasoning, and lower costs through more efficient token usage. Flash Lite gains stronger multimodal and translation skills, while Flash improves tool use for complex tasks.

Key details:

  • Gemini 2.5 Flash-Lite is amongst the cheapest models and now further reduces output token use by 50%, Flash by 24%

  • Flash boosts coding performance by 10% on SWE-Bench Verified

  • With “thinking” enabled, Flash delivers higher-quality answers while using fewer tokens

  • New “-latest” alias in the API names (e.g., gemini-flash-latest) removes the need to update with each preview

Why it matters: Gemini Flash has quietly become the backbone for software applications, with developer usage increasing 10x in late 2024. Its economics with low token cost and massive context windows power many AI infrastructures today. Control of this runtime layer may matter more than raw benchmarks, as it decides who owns the default AI stack for real-world applications outside the chatbot interface.

FROM OUR PARTNERS

A Smarter Way to Read the News

Seeking impartial news? Meet 1440.

Every day, 3.5 million readers turn to 1440 for their factual news. We sift through 100+ sources to bring you a complete summary of politics, global events, business, and culture, all in a brief 5-minute email. Enjoy an impartial news experience.

OPENAI

AI models tested on real-world jobs with GDPval

The Summary: OpenAI introduced GDPval, a new benchmark that measures AI performance on real-world tasks across 44 professional occupations. Instead of abstract tests, GDPval uses real deliverables such as legal briefs, engineering plans, and research reports to see how well models perform.

Key details:

  • GDPval tasks cover healthcare, finance, government, manufacturing

  • Each task represents several hours of skilled work

  • Tasks include multimedia outputs such as preparing documents, spreadsheets, and presentations, graded blindly by domain experts

  • Claude Opus 4.1 leads, matching experts in 47.6% of cases

Why it matters: GDPval offers a more grounded way to track AI progress on specific real-world tasks, showing where models can assist with well-defined assignments. By focusing on specific tasks, the benchmark points toward a collaboration model. Over time, GDPval could help decide where AI adds the most value without displacing the human core of work.

QUICK NEWS

Quick news

TOOLS

🥇 New tools

That’s all for today!

If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/