GPT-5 Is Finally Here, And The Wait Was Worth It

OpenAI releases GPT-5, its new default model in ChatGPT, replacing GPT-4o and earlier versions. The rollout starts today for Plus, Pro, Team, and Free users, while Pro subscribers gain unlimited use and an enhanced GPT-5 Pro edition.

GPT-5 combines a fast-response model, a deeper “thinking” engine, and a real-time router, cutting factual errors by roughly 45% versus GPT-4o and 80% versus OpenAI o3. It outperforms prior models in coding, writing, health queries, and sets new top scores across math, multimodal, and software-engineering benchmarks.

External reviewers preferred GPT-5 in 67.8% of 1,000 economically valuable reasoning tasks and logged 22% fewer major errors. OpenAI’s internal tests show GPT-5 equals or surpasses human experts in about half of cases across 40 professional occupations.

Meta, Alphabet, Microsoft and Amazon have driven combined capex to a record $102.5 billion in their latest quarters, eclipsing the $6.7 billion spent by Apple, Nvidia and Tesla.

Private credit lenders now back roughly $50 billion in data-center-linked securities and finance chip purchases for firms like CoreWeave and Fluidstack.

Boston Fed researchers warn that highly correlated defaults in private-credit portfolios create a tail risk for banks despite the loans’ senior status.

Google has added a guided learning mode to its Gemini AI that coaches users through problems with step-by-step questions instead of simply giving answers. CEO Sundar Pichai says the feature was shaped with input from students, educators, researchers, and learning experts.

Responses can include images, videos, and interactive quizzes, and students aged 18 and older in the US, Japan, Indonesia, Korea, and Brazil can claim a free year of the Gemini AI Pro Plan if they enroll by October 6. Google also committed $1 billion over three years to support American education through AI literacy initiatives, research funding, and cloud resources.

The launch follows OpenAI’s similar study mode for ChatGPT, highlighting an accelerated push by AI firms into the classroom. The article notes these tools aim to shed chatbots’ cheating reputation and questions whether students will opt for guided learning over quick answers.

Windsurf scrapped a $2.85 billion funding round, saw a $3 billion sale to OpenAI collapse, and ultimately fractured—its founders and top staff joining Google while key shareholders received a $2.4 billion payout.

Insiders say code-generation products often run on “very negative” gross margins because the newest, most capable LLMs cost more to operate than users pay.

Rising LLM charges force startups to either build proprietary models or push fees onto customers, a tension that sparked backlash when Anysphere added surprise surcharges for heavy Cursor users.

Google DeepMind CEO Demis Hassabis warns that AI’s impact will be “10 times bigger than the Industrial Revolution” and maybe 10 times faster. He places the arrival of human-level artificial general intelligence within the next five to ten years.

Fresh from winning a Nobel Prize in chemistry for AlphaFold’s protein-structure breakthrough, Hassabis says DeepMind now powers every corner of Google’s AI push, from search summaries to the Gemini assistant. He adds that Meta, Microsoft, Amazon and Apple are dangling $100 million pay packages to lure top researchers away from DeepMind.

Hassabis couples his promise of “radical abundance” with blunt concerns over soaring datacenter energy needs, deepfakes and jobs already being automated. The interview captures a tech industry racing forward while regulators and societies struggle to keep pace.