- The Summary AI
- Posts
- OpenAI Launches o3-Pro
OpenAI Launches o3-Pro
PLUS: ByteDance Beats Google Veo 3
The Summary AI
June 13, 2025
Welcome back!
OpenAI just flipped the switch on its most advanced model yet, O3-Pro, now live for ChatGPT Pro users and developers, shifting from conversationist to analyst who can generate strategic plans from complex raw data. Think company roadmaps, code audits, and decision-making workflows. Let’s unpack…
Today’s Summary:
🔥 OpenAI launches o3-Pro
🎥 ByteDance video AI beats Google Veo 3
🌪️ Google DeepMind advanced weather models
🤖 Meta V-JEPA 2 learns physics
⏳ OpenAI delays open-source release
⚖️ Disney & Universal sue Midjourney
🧠 Altman says superintelligence work started
🛠️ 2 new tools
TOP STORY
OpenAI launches o3-Pro
The Summary: OpenAI has rolled out o3-pro to ChatGPT Pro users and API developers, replacing o1-pro. Priced 87% cheaper than o1-pro, it can perform complex tasks like planning, coding, and data analysis, when given rich context. It’s slower as it thinks longer, but users say it delivers sharper and more actionable results.
Key details:
Outperforms o1-pro and o3 in evals: 93% on AIME math, 84% on GPQA Diamond science, 2748 on Codeforces
First reviewers report o3-pro is preferred over o3 for clarity, planning, and instruction-following
Unlike chat-optimized models, o3-pro acts more like an analyst for slow tasks taking 10-20 minutes to solve, not for casual conversation
API cost decreased to $20 per million input tokens and $80 for output, down from $150 and $600 for o1-pro
Now live for ChatGPT Pro and Team users. Enterprise and Edu will get it next week
Why it matters: o3-pro shows where large models are headed: deeper, slower, more deliberate thinking. Its strength lies in being able to digest massive context and return structured, actionable insight. This model changes the dynamic from asking AI for a quick conversational answer to trusting it with a bigger problem and expecting a whole plan.
BYTEDANCE
Seedance video AI challenges Google Veo 3
The Summary: ByteDance unveiled Seedance 1.0, a high-quality AI video generator that’s already outperforming Google Veo 3 in benchmarks. The model creates full-scene, multi-shot videos from text or image prompts, with a focus on consistency, motion quality, and stylistic control.
Key details:
Generates 5 seconds of 1080p video in 40 sec
Ranked above Veo 3, Sora, and Kling 2 in Artificial Analysis leaderboards
Native multi-shot storytelling keeps subjects, scenes, and camera angles coherent
Lacks audio generation, Veo 3 still leads in that area
Why it matters: Veo 3 grabbed headlines for its exceptional video quality, but Seedance is now delivering the same quality, minus the sound. What makes this disruptive is that ByteDance owns TikTok, so once they integrate it into that platform, every swipe could at some point trigger personalized renderings, not just play existing clips. That could introduce AI-native media feeds. Will the creator economy become a prompt economy?
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.
Google AI predicts storm paths better than physics models
The Summary: Google DeepMind has launched an AI system to predict tropical cyclones more accurately and much faster than traditional models. The tool is already being used in real-time forecasts by the U.S. National Hurricane Center. Users can explore the AI’s predictions on a new interactive tool that compares AI scenarios to physics-based models.
Key details:
Improves ECMWF’s ENS 5-day storm track predictions by 87 miles
Beats NOAA’s top tool on storm intensity forecasting
Forecasts run in about 1 minute, compared to hours for traditional models
Weather Lab tool lets users explore weather scenarios
Why it matters: DeepMind’s model leapfrogs a decade of gradual advances that physics models would have required years to match. For forecasters and disaster responders, the extra lead time and accuracy could directly save lives. Private-model integration into public systems adds a critical new layer of infrastructure.
QUICK NEWS
Quick news
Meta V-JEPA 2 learns real-world physics to navigate new environments
Disney & Universal sue Midjourney for copyright infringement
Sam Altman says work on “Superintelligence” has begun
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/