OpenAI news and ChatGPT upgrades: GPT-5, Agents, Open Source and a massive funding round
GPT-5 is now the default in ChatGPT, replacing GPT-4o and o-series models. It’s faster, more accurate, better at reasoning, and far less likely to hallucinate. The new “thinking” mode dynamically decides when to reason deeply, with a Pro tier (GPT-5 Pro) offering extended reasoning for the most complex tasks. Gains are especially strong in coding, health, and multimodal reasoning. Though, this isn’t AGI yet.
Agents mark ChatGPT’s move from “tell me something” to “do it for me.” The Agent can browse, execute code, fill forms, integrate with services like Gmail and GitHub, and complete multi-step workflows. Here is everything you need to know about it!
Open Source: OpenAI also released two open-weight models — gpt-oss-120B and gpt-oss-20B — under Apache 2.0. They match or beat many proprietary models on reasoning benchmarks, and can run locally (the smaller one with 16 GB of RAM). This is a major step toward more privacy — and a sign OpenAI wants to influence the open-source ecosystem rather than leave it to competitors.
Funding: All this comes as OpenAI closes an $8.3 billion round (part of a $40 billion raise), with ARR jumping to $13 billion and projected to hit $20 billion by year-end. Paid business ChatGPT users are now over 5 million. The round was oversubscribed 5×, showing investors still see OpenAI as the leader despite intense competition from Anthropic, Google, and Meta.
Gemini 2.5 Deep Think: Google’s slow but sharp “math medalist”
Google has launched Gemini 2.5 Deep Think — but initially only for its $250/month AI Ultra subscribers. It’s designed for complex reasoning, with longer “thinking time” and parallel hypothesis testing.
The model outperforms OpenAI o3 and Grok 4 in several benchmarks, scoring 34.8% on Humanity’s Last Exam and earning a bronze medal in the 2025 IMO (International Mathematical Olympiad). A special internal version even won gold.
The catch? Answers can take minutes, there’s a daily cap on queries, and it’s locked behind one of the priciest AI subscriptions in the market.
This is clearly Google aiming for the “elite problem solver” segment — the question is whether enough people are willing to pay for it.
AI in the classroom: 400,000 teachers to lead the AI shift
OpenAI and the American Federation of Teachers are launching the National Academy for AI Instruction, a 5-year program to train 400,000 K-12 educators in AI use and policy leadership.
OpenAI is contributing $10 million (funding + in-kind resources), joined by partners like Microsoft and Anthropic. The initiative will offer workshops, hands-on training, and AI tool access, with a flagship facility in New York and national expansion planned.
My view: This is the right direction. If AI is going to transform education, it should be shaped by teachers, not imposed on them. The key will be ensuring the tools enhance learning rather than replace it. AI Education is becoming a must, like learning how to read and count.
DOGE’s bold move: AI to delete 100,000+ US regulations
The Department of Government Efficiency (DOGE) is using AI to scan 200,000 federal rules and create a “delete list,” aiming to cut 50% of regulations by January. Some agencies are reportedly using AI to write all their deregulations.
From my perspective, this is both fascinating and risky. If France — with its 400,000+ business regulations — tried this, it could unleash billions in economic gains… or plunge the system into legal chaos.
AI generated content: demonetization as ultimate weapon?
YouTube is tightening monetization rules to target “mass-produced, repetitious” AI-generated videos — the stock footage + robotic voiceover spam flooding the platform.
This isn’t a ban on AI, but a call for quality. Platforms like Udemy and Kindle Direct Publishing are making similar moves, rejecting low-effort AI output. The bar is rising: assisted creation is fine, low-value auto-content is not.
