AI News - Saturday, 7 November 2026 - OpenAI $11B Raise, Anthropic IPO Resistance, Inkling 2 Small
In a Nutshell
Capital is repricing the AI trade while capability keeps shipping. OpenAI reportedly raised $11 billion at a $1.25 trillion valuation even as Anthropic's IPO met buyer resistance and enterprise AI spend flattened. Open-weight releases from Thinking Machines, Ricursive and Baidu narrowed the gap, and GPT-6 Cyber turned automated vulnerability discovery into a product.
5-minute AI news update - 7 November 2026
In this AI News
OpenAI reportedly adds $11 billion at a $1.25 trillion valuation
[Funding] The raise lands during a broad tech selloff and while rival Anthropic's IPO meets investor pushback on price. Private capital is still clearing at record levels even as public markets turn cautious, a gap worth watching for anyone who plans AI budgets around vendor stability. Source: TechCrunch
Anthropic's $1.5 trillion IPO meets investor resistance as tech stocks sell off
[Funding] Buyers are reportedly anchoring nearer $1.1 trillion, a wide spread for a deal of this size. A price below expectations would reset the valuation narrative for every private AI lab and hand enterprise buyers a stronger position in multi-year contracts. Source: The Decoder
Thinking Machines ships Inkling 2 Small, claiming top western open-weight model
[Models] It is a 594 billion parameter mixture-of-experts model from Mira Murati's team, positioned as the strongest open-weight release outside China. The better agentic scores come with sharply higher token use, so cost per completed task matters more than the leaderboard line. Source: The Decoder
Ripple-65B Preview lands as a fully Apache 2.0 US open model
[Models] Ricursive Intelligence says the 65 billion parameter model tops open US models on TerminalBench 2.1 coding at $32 million total R and D. It trails Inkling 2 Small on other benchmarks, which shows how narrow benchmark wins have become. The permissive licence is the more durable signal for institutions that need to self-host. Source: The Decoder
OpenAI's GPT-6 Cyber finds vulnerabilities, writes exploits and proposes patches
[Models] OpenAI says the model surfaced 81 previously unknown Firefox flaws and produced 948 working Linux kernel exploits, far more than its predecessor. Defensive patch cycles will compress, and every IT team should assume automated vulnerability discovery is now cheap for attackers as well. Source: The Decoder
Gemini Omni 1.2 Flash doubles clip length to 15 seconds and cuts price 30 percent
[Models] Google DeepMind's video model now costs about $0.07 per second at 720p and supports step by step conversational edits. Falling video generation cost changes the economics of course production and campaign assets, and it pushes institutions to set a synthetic media policy before staff adopt the tools informally. Source: Google
Baidu releases ERNIE 5.5 Flash Base, a 280 billion parameter open model
[Models] The mixture-of-experts model was pretrained on 33.5 trillion tokens and shipped deliberately without fine-tuning so developers can adapt it themselves. Chinese labs keep releasing strong weights openly while US frontier labs stay closed, which shapes where cost-sensitive deployments end up. Source: The Decoder
METR finds Anthropic may have crossed its own AI acceleration warning threshold
[Research] The nonprofit spent eleven weeks inside Anthropic's research data and concluded the pace of model development roughly doubled, with about 80 percent of the speedup coming from AI-assisted research. Anthropic redacted the most informative evidence, so the finding rests on METR's judgment more than ideal. Safety commitments need independent audit access to stay credible. Source: The Decoder
A humanoid robot cleaned an apartment, and the review reads like a preview
[Research] MIT Technology Review tested a commercial humanoid cleaning service in San Francisco and found it close to useful rather than a novelty. Service robots entering domestic work is the clearest sign yet that embodied AI is moving from demonstration to contract. Source: MIT Technology Review
Crusoe sells its troubled Wyoming data center campus to Google
[Industry] The transfer shows how hard it is to build very large AI campuses on schedule and on budget. Google absorbing a stalled project concentrates compute capacity further among hyperscalers, which matters for any institution negotiating inference capacity or regional hosting. Source: TechCrunch
Meta adds Spaces so Muse agents can be organised by task
[Tools] Spaces lets users create separate working contexts for the same assistant instead of one long thread. Context separation is becoming the main design battleground for agents, and it maps directly onto how staff separate teaching, research and administrative work. Source: TechCrunch
Perplexity launches always-on Teammates agents on dedicated virtual machines
[Tools] The agents keep working through long-running projects on their own machines, priced at $200 per month on the Max plan. That sits above comparable OpenAI and Anthropic offers and the always-on category is crowding fast. Buy on measured task completion, not on the demo. Source: The Decoder
Google tells schools they can skip Gemini but not AI search
[Education] Schools can opt out of Gemini classroom tools, yet Google keeps AI overviews in search on by default. The choice is narrower than it looks, and it is a live issue for any institution that assumes a platform opt-out removes AI from the student experience. Source: Ars Technica
Survey finds 92 percent of HR believe staff can be reskilled for AI roles
[Education] Confidence in reskilling is high, yet roughly half of companies still hire externally for the same roles. The gap is usually assessment and time, not belief. Institutions that build internal certification and evidence of applied work close it faster than those that buy talent. Source: ZDNET
Wired: free Chinese open-weight models are nearing a cyber capability milestone
[Research] Openly downloadable models are approaching the point where they can support meaningful offensive security work without a paid API. That removes the last cost barrier for misuse and puts pressure on institutions to monitor what models run on their own hardware. Source: Wired
Hundreds of exposed Artex servers put AI hacking tools on the open internet
[Policy] Researchers found hundreds of internet-facing servers hosting the Artex penetration testing tool, reachable by anyone. AI-assisted attack tooling is leaking out of controlled environments, so external attack surface review is no longer optional. Source: TechCrunch
Ramp data shows AI usage rising while enterprise AI spending stalls
[Industry] Usage keeps climbing but spend has flattened as price cuts and cost controls take hold among the largest buyers. That contradicts the hypergrowth assumptions behind current valuations. Budget planning should assume cheaper tokens each quarter, not a fixed per-seat cost. Source: TechCrunch
US Department of Justice creates a Super Intelligence Task Force
[Policy] The unit is framed around preventing the technology from being manipulated by bad actors at home and abroad. It adds a law enforcement layer on top of existing AI regulators, and early signals from such units tend to shape procurement expectations before they shape formal rules. Source: Engadget
The world of AI is evolving at full speed.
Become a Fellow at university-365.com
Become Superhuman... In a world of AI... Prompt Smart, Prompt UP!
Comments