OpenAI leans toward waiting until next year for IPO, NYT reports ↗
OpenAI is reportedly weighing a slower IPO path, with advisers floating a choice: wait for a $1 trillion valuation, or go sooner at a lower number. Sam Altman, by the sound of it, did not warm to the discounted version.
Separately, the Trump administration asked OpenAI to stagger GPT-5.6 access over security concerns. The model is set for a limited preview, with access approved customer by customer - unusually hands-on for a supposedly “hands-off” AI era. (Reuters)
After Anthropic shutdown, China's Z.ai closes frontier gap as it plans dual listing ↗
Z.ai says it will use domestic listing proceeds to chase AGI after GLM-5.2 landed close to top U.S. models on public benchmarks. That is not a small footnote - it is more like a chair scraping loudly in the frontier-model room.
The model has 750 billion parameters and a 1-million-token context window, with adaptation for domestic Chinese chip infrastructure. Z.ai’s market value crossed HK$1 trillion after a massive stock run, because apparently AI gravity got turned off again. (Reuters)
ChatGPT - Release Notes ↗
OpenAI made Codex Remote generally available on all ChatGPT plans. Users can start or continue work on a connected Mac or Windows host from mobile, then review progress and approve actions from their phone. Handy, faintly uncanny, and genuinely practical.
There is also a new DigitalOcean Droplet Workspace plugin, letting Codex provision a Droplet, configure SSH, and connect it as a remote workspace. A tiny devops gremlin in your pocket, more or less. (OpenAI Help Center)
Patronus AI lands $50M to build ‘digital worlds’ that stress-test AI agents ↗
Patronus AI raised $50 million to build simulated environments where AI agents can be tested before being trusted with live operational tasks. Benchmarks look polished, sure, but they do not prove an agent can reliably book the trip or run the finance workflow without chewing the furniture.
Its “digital world models” recreate websites and internal systems so agents can be pushed through gnarly scenarios, rewarded for success, and penalised for errors. The startup says revenue has grown 15-fold, which explains the investor stampede. (TechCrunch)
Databricks’ former AI chief thinks he can cut AI’s power bill by 1,000x ↗
Unconventional AI, led by former Databricks AI chief Naveen Rao, unveiled Un-0, an image-generation model built around a simulated oscillator-based architecture. Big claim, small name - very AI-startup-coded.
The pitch is that this architecture could eventually slash inference power use by up to 1,000 times. For now it is running in software simulation, so the real chip test still matters, but the ambition is properly unhinged. (TechCrunch)
🛒 IndiaMART doubles down on AI to curb fake listings, improve buyer interaction ↗
IndiaMART plans to double AI spending every six months as it tries to catch fake listings and improve content checks. The marketplace handles everything from phone chargers to industrial machinery, which sounds like an AI moderation gym with no closing time.
The company is using pattern matching to detect proxy seller accounts and real-time voice-to-text to process buyer requests faster. It says it connects about 600 buyers with suppliers every minute. (Reuters)
💾 Micron overtakes Meta, Tesla in market value amid relentless AI infrastructure demand ↗
Micron briefly pushed past Meta and Tesla in market value after a strong forecast reignited the AI memory-chip trade. Memory is suddenly not the dull cupboard in the data-centre kitchen - it is the whole pantry.
Shares jumped 18.4%, and the company disclosed $22 billion in customer commitments to lock in memory-chip supply. That says a lot about how desperate the AI infrastructure race has become. (Reuters)
🇪🇺 Anthropic hires Orange's AI chief amid Europe push ↗
Anthropic hired Steve Jarrett, Orange’s chief AI officer, as it keeps expanding in Europe. He will be based in Paris and focus on adapting Anthropic’s products for European and African markets.
The move fits a broader international push: Anthropic recently opened a Milan office and plans to triple its international workforce. Claude is packing a suitcase, basically. (Reuters)
FAQ
Why might OpenAI wait until next year for an IPO?
OpenAI is reportedly considering a slower IPO path because advisers have discussed whether waiting could support a much higher valuation. The article says one option involved waiting for a possible $1 trillion valuation, while another would mean going sooner at a lower figure. Sam Altman reportedly did not favour the lower-valuation route.
What does staggered GPT-5.6 access mean for customers?
Staggered GPT-5.6 access means the model would not become broadly available to everyone at the same time. According to the article, access would be approved customer by customer because of security concerns raised by the Trump administration. This suggests a more controlled rollout, especially for users or organisations seeking early preview access.
Why is China’s Z.ai being watched in frontier AI?
Z.ai is being watched because its GLM-5.2 model reportedly performed close to top U.S. models on public benchmarks. The article also notes that the model has 750 billion parameters, a 1-million-token context window, and adaptations for domestic Chinese chip infrastructure. That combination makes it relevant to the frontier AI race.
What is Codex Remote in ChatGPT used for?
Codex Remote lets users start or continue coding work on a connected Mac or Windows host from mobile. Users can review progress and approve actions from their phone, making it valuable for remote development workflows. The article also mentions a DigitalOcean Droplet Workspace plugin for provisioning and connecting a remote workspace.
Why are digital worlds important for testing AI agents?
Digital worlds give companies a safer way to test AI agents before letting them handle live operational tasks. Patronus AI is building simulated websites and internal systems where agents can be pushed through difficult scenarios. This helps expose errors in workflows such as booking, finance tasks, or other multi-step actions before they affect live systems.
How does AI infrastructure demand affect companies like Micron?
AI infrastructure demand can increase the value of companies that supply critical data-centre components. In the article, Micron briefly overtook Meta and Tesla in market value after a strong forecast and disclosed $22 billion in customer commitments for memory-chip supply. This shows how important memory has become in the AI infrastructure race.