GPT-4.5 Is Here... But There's a Catch

PLUS: Tencent vs. DeepSeek

Author

Jack Comey
February 28, 2025

Welcome to Friday’s Newsletter

In today’s scoop 🍨 

  • GPT-4.5: Smarter, Costlier & GPU-Starved

  • Tencent's AI Turbo: Speed Over Smarts? 

  • Sora Finally Lands in Europe

  • 3 Trending AI Tools

🚀 GPT-4.5: Smarter, Costlier & GPU-Starved

OpenAI just dropped GPT-4.5, its latest and largest AI model, but there’s a catch—it's so massive that OpenAI has literally run out of GPUs. Yep, the company behind the world’s most powerful AI is now scrambling for chips like gamers on a Black Friday sale.

🤖 What's New in GPT-4.5?

GPT-4.5, code-named Orion, is a significant upgrade, but not quite the revolution some hoped for. Here’s what’s changed:

  • 🧠 More Knowledge, Fewer Hallucinations: GPT-4.5 has a deeper understanding of the world and fabricates less nonsense.

  • 💻 Sharper Writing & Coding: It’s better at recognizing patterns and solving complex problems, making it a top-tier AI for programming and content creation.

  • 🎙️ Smoother Conversations: OpenAI claims it has a "warmer, more intuitive" personality, making interactions feel more natural.

But despite these improvements, it’s not a “frontier AI”—meaning it won’t be setting new benchmarks against reasoning models from competitors like Anthropic and DeepSeek.

💸 The Cost? Eye-Watering

GPT-4.5 is expensive. OpenAI is charging $75 per million input tokens and $150 per million output tokens—30x the cost of GPT-4o. That’s like switching from budget coffee to a gold-plated espresso machine.

🛑 The GPU Crisis

Sam Altman, OpenAI’s CEO, admitted that they're completely out of GPUs, which is why access is rolling out in waves. ChatGPT Pro users get it first, followed by Plus and Team users next week.

🌍 The Bigger Picture

With AI scaling laws showing diminishing returns, OpenAI is now pivoting toward reasoning models, merging the GPT series with its "o" models for future AI breakthroughs. So while GPT-4.5 might not be a game-changer, it’s likely a stepping stone to GPT-5 and beyond.

The takeaway? More power, higher costs, and a reminder that even OpenAI isn’t immune to supply chain nightmares.

🧠Tencent's AI Turbo: Speed Over Smarts? 

Tencent just threw a turbocharged wrench into the AI race with its latest model, Hunyuan Turbo S, boasting it can outpace DeepSeek-R1 in response time. The move signals a growing trend in AI—faster, cheaper, and (hopefully) smarter. But does speed trump intelligence? Let’s dig in.

✨ Fast & Furious: AI Edition

  • 🏎️ Lightning-Fast Replies: Turbo S delivers responses in under a second, a stark contrast to "slow-thinking" models like DeepSeek-R1.

  • ⏱️ Speed vs. Depth: Unlike DeepSeek’s deep-reasoning approach, Tencent's model prioritizes quick, intuition-based answers before layering in logic.

  • 💡 Tech Upgrade: A hybrid Mamba-Transformer architecture makes this possible, cutting computational complexity and slashing costs.

🔎 Why Does This Matter?

  • 🔄 AI Arms Race Intensifies: Tencent joins Alibaba, Baidu, and ByteDance in the battle to dethrone DeepSeek, which has become China’s AI golden child.

  • 💰 Cheaper to Deploy: Reduced training and deployment costs mean Turbo S could be widely adopted, especially in enterprise and WeChat integrations.

  • ⚙️ OpenAI Who? DeepSeek shocked Silicon Valley by competing with OpenAI and Meta’s best models—and now, Tencent is using it as a benchmark.

🔑 Takeaway

The question isn’t just who has the best AI but who has the fastest, most scalable, and cost-effective model. Tencent's Turbo S prioritizes speed, but will it match DeepSeek's reasoning prowess? The AI race isn’t slowing down—and neither is Tencent.

🎥 Sora Finally Lands in Europe

The wait is over—OpenAI’s AI video generator, Sora, has finally made its way to the UK and Europe. After months of regulatory delays, ChatGPT Plus and Pro users in the region can now create videos straight from text prompts, making your wildest cinematic dreams (or weirdest AI hallucinations) a reality.

🚀 What’s New?

  • Sora can generate 1080p, 20-second videos in multiple aspect ratios.

  • It comes with tools like Remix (swap or remove elements), Re-cut (extend scenes), and Loop (create seamless repeats).

  • ChatGPT Plus users get 50 videos per month (at 720p, max 5 sec), while Pro users can go all out with 500 fast generations and unlimited slow ones.

🔎 Why Does This Matter?

  • OpenAI has been working with select European artists to test Sora, including UK-based creators who’ve already made AI-generated short films.

  • The Discord Sora community is launching, featuring office hours and masterclasses for creators.

  • Despite its power, Sora still struggles with object permanence (floating hands, anyone?) and anatomical realism—expect some cursed-looking humans.

💡 The Big Picture

With competitors like Google’s Veo 2, Runway’s Gen-3 Alpha, and Adobe’s Firefly Video, the AI video race is heating up. But for now, Sora is finally here in Europe—so get ready to turn your text prompts into Hollywood (or horror) magic.

  • 🎙️ Octave TTS - The first LLM-powered text-to-speech that understands meaning, letting you create AI voices and control emotional delivery with descriptive prompts for lifelike expression.

  • 🗣️ Rapport AI-Driven Avatars- Bring AI to life with real-time, voice-driven digital characters, now featuring animated laughter, breathing, and an Unreal Engine plug-in for immersive conversations.

  • 📚 Deep Review by SciSpace - AI co-scientist that conducts systematic literature reviews, refining searches, analyzing vast datasets, and delivering the most relevant academic papers.

📩 Before you head into the weekend…

We’re giving away a free expert prompting guide to help you save hours every week—whether in work or daily life. Want a copy? Just hit reply and let us know what you think of the newsletter, and we’ll send it your way! 🚀

And if you know someone who’d love this, pass it along and help grow the community!

Enjoy your weekend, and we’ll see you Monday! 🎉👋