🚀 OpenAI Launches GPT-5.2

PLUS : Runway 4.5 New Video Leader

Author

The Summary AI
December 12, 2025

In partnership with

Welcome back!

OpenAI is back in the race with GPT-5.2, a strong new release that narrows the gap with Gemini 3. The update pushes reasoning and coding benchmarks to new highs, adds a 400k-token API context window and sharpens tool accuracy. Let’s unpack…

Today’s Summary:

  • 🚀 OpenAI launches GPT-5.2

  • 🎬 Runway Gen-4.5 surpasses Google Veo

  • 🔮 Demis Hassabis forecasts AI in 2026

  • 💡 Alibaba Qwen3 Omni-Flash upgrade

  • 🗣️ ChatGPT Voice integrated in chat

  • 🧩 Google adds Workspace AI Studio

  • 🛠️ 2 new tools

TOP STORY

OpenAI launches GPT-5.2

The Summary: OpenAI has released GPT-5.2, a substantial leap in performance and reliability over GPT-5.1. The model posts new highs in reasoning, coding, and factual precision, with large boosts across benchmarks such as GDPval, SWE-Bench Pro, and ARC-AGI-2.

Key details:

  • Factual error rate is down 30%, with tool use accuracy at 98.7%

  • Improvements in web design and visual understanding

  • New 400k-token context window in the API

  • On ARC-AGI-2 hard reasoning test, GPT-5.2 Thinking jumped to 52.9% from 17.6% in GPT-5.1, one of the largest gains reported

  • Rolling out now in ChatGPT Free, Plus, Pro, Business, and Enterprise

Why it matters: GPT-5.2 lands amid an internal “code red” at OpenAI to sharpen ChatGPT performance after Gemini 3 had reclaimed the industry’s #1 position. GPT-5.2 closes much of that gap, but doesn’t clearly take back the crown. Early benchmarks show near parity, matching Gemini 3 in some tasks, while trailing in others. Still, it puts OpenAI firmly back in the race.

FROM OUR PARTNERS

Turn Any Workflow into a Tutorial in Seconds

Effortless Tutorial Video Creation with Guidde

Transform your team’s static training materials into dynamic, engaging video guides with Guidde.

Here’s what you’ll love about Guidde:

1️⃣ Easy to Create: Turn PDFs or manuals into stunning video tutorials with a single click.
2️⃣ Easy to Update: Update video content in seconds to keep your training materials relevant.
3️⃣ Easy to Localize: Generate multilingual guides to ensure accessibility for global teams.

Empower your teammates with interactive learning.

And the best part? The browser extension is 100% free.

RUNWAY AI

Runway Gen-4.5 moves ahead of Google Veo

The Summary: Runway released Gen-4.5, a new text-to-video model focused on tight prompt control, stable motion, and realistic physics. The update adds longer multi-shot generation with character continuity and native audio. Runway also introduced GWM-1, a general world model that runs in real time, designed for interactive simulation and robotics.

Key details:

  • Keeps Gen-4 speeds and pricing while adding native audio, multi-shot editing, and one-minute videos with character consistency

  • Leads Artificial Analysis’ Text-to-Video benchmark ahead of Google Veo 3 and OpenAI Sora 2 Pro

  • Built for production use by studios, brands and agencies, while also keeping a free plan and low-cost tiers for individuals

  • GWM-1 runs at 24 fps and 720p, includes a Robotics version

Why it matters: Gen-4.5’s gains in temporal stability make longer sequences usable, lowering the cost of some video production workflows. That consistency is a prerequisite for simulation and safe robotics testing.

FROM OUR PARTNERS

Smarter Prompts for Faster Results

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.

GOOGLE

Demis Hassabis’ AI predictions for 2026

The Summary: DeepMind CEO Demis Hassabis predicts that the next wave of AI progress in 2026 will center on multimodal understanding, interactive video worlds, and more reliable agents. He says AI agents will soon be able to complete complex tasks with limited supervision.

Key details:

  • DeepMind’s Nano Banana model can generate accurate infographics by understanding structure and meaning in visuals

  • AI interprets video scenes at a symbolic level rather than purely visual clues

  • Agents could handle full task delegation in 2026

Why it matters: The claim here is about reliability. Models already see, hear, and talk, but they may soon stay coherent long enough to be reliable. World models matter because they give agents a memory of consequences. Reliable agents could change software economics by shifting value from features to measurable outcomes.

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/