OpenAI released GPT-5.2 on Thursday, which it deems its “most capable model series yet for professional knowledge work,” with performance gains across writing, coding, and reasoning. The launch comes just days after Sam Altman declared a “code red” within the company, urging engineers to improve ChatGPT to compete with rivals who had recently stepped up their game.
The company wrote:
“Already, the average ChatGPT Enterprise user says AI saves them 40–60 minutes a day, and heavy users say it saves them more than 10 hours a week. We designed GPT‑5.2 to unlock even more economic value for people; it’s better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long contexts, using tools, and handling complex, multi-step projects.”
So far, after using GPT-5.2 for the past few days, I've got to agree with them. It feels like a major step up for some tasks over the previous model. (Although creating images with GPT-5.2 still feels slower than molasses going uphill in January — so not all areas of the model were improved upon.)
OpenAI says it concentrated on bringing improvements to general intelligence, long-context understanding, agentic tool-calling, and vision, aimed at making the model better at executing complex, real-world tasks.
Google wasn't letting OpenAI have all the fun last Thursday…
On the same day as GPT-5.2 dropped, Google released a “reimagined” version of its research agent, Gemini Deep Research, based on its Gemini 3 Pro model.
The company says:
“Gemini Deep Research is an agent optimized for long-running context gathering and synthesis tasks. The agent’s reasoning core uses Gemini 3 Pro, our most factual model yet, and is specifically trained to reduce hallucinations and maximize report quality during complex tasks. By scaling multi-step reinforcement learning for search, the agent autonomously navigates complex information landscapes with high accuracy.”
Google says it will eventually integrate this new deep research agent into Google Search, Google Finance, Gemini App, and NotebookLM.
Last but not least, Meta is working on a new closed-source model named “Avocado”.
The model is being worked on inside of “TBD,” which is a smaller group within Meta's AI Superintelligence Labs, and marks a significant shift away from the company's previous AI approach of building open-source LLMs. Last year Mark Zuckerberg famously said “fuck that” to closed-source models, however he's since committed hundreds of billions of dollars to AI acquisitions, data center development, and onboarding talent for his Superintelligence Labs, so perhaps his tune has changed.
As for its Llama model, the newest version has been delayed for months, and The New York Times reported earlier this year that the company's newish Chief AI Officer Alexandr Wang and other execs had discussed abandoning it altogether. Zuckerberg also made comments earlier this year that the company may need to shift its approach to open-source to mitigate potential safety risks (such as the safety of Meta's stock price).
