AI News - Thursday, 24 September 2026 - GPT-6 Sol and Luna Pricing, Claude Opus 5.5 Cost, Meta Muse Hardware
In a Nutshell
OpenAI and Anthropic shipped in the same week, with GPT-6 Sol and Luna and Claude Opus 5.5 landing as cheaper frontier tiers. Meta pushed Muse from app to hardware at Connect 2026, while Anthropic's biology lab reported agents discovering a novel enzyme system. Regulators kept moving: a US bill would ban superintelligence, Australia disclosed an OpenAI agent breach, and AI safety dominated the UN.
5-minute AI news update - 24 September 2026
OpenAI launches GPT-6 Sol and Luna with lower cost and fewer mistakes
[Models] OpenAI's two new models are positioned as a cheaper frontier tier cut from the same cloth as Astra. For U365 this is the clearest signal yet that frontier capability is commoditising, so curriculum and platform choices should assume model swapping rather than vendor lock-in. Source: TechCrunch
Anthropic and OpenAI ship cheaper models days after calling for a slowdown
[Models] Anthropic unveiled Claude Opus 5.5 while OpenAI introduced GPT-6 Sol and GPT-6 Luna, both framed on price rather than raw capability. The timing undercuts the safety-first tone both labs used at the UN this week. Source: CNBC
Google releases Gemini 3.8 text-to-speech models for expressive audio
[Models] Gemini 3.8 Flash-Lite TTS and 3.8 Flash TTS are Google's most expressive audio models to date. Voice output at this quality changes what a course interface or campus assistant can sound like. Source: Google Blog
Inception's Mercury 2.5 hits 770 output tokens per second
[Models] Speed at this level moves latency out of the design constraint for agentic workflows. Fast, cheap inference is what makes always-on tutoring and multi-step campus agents viable. Source: Artificial Analysis
Anthropic says Claude agents discovered a novel enzyme system
[Research] Early results from Anthropic's life sciences lab describe an enzyme system whose function is still unknown. It is the first real output from the argument that AI agents, not just AI models, drive scientific discovery. Source: Anthropic
arXiv secures multiyear funding to become an independent nonprofit
[Research] Three philanthropic organisations committed multimillion-dollar support to arXiv's independence. arXiv is the substrate much of applied AI research, including U365's, is built on, so its funding stability matters. Source: arXiv Blog
Apple releases LensVLM, compressing long context into images
[Research] LensVLM-9B stores long context as images and expands only the pages it needs. If it holds up, context compression of this kind lowers the cost of long document work common in academic settings. Source: Hugging Face
ChatGPT mobile app gains voice-driven agentic task features
[Tools] Pro and Plus users can now run agentic tasks from the phone's Work tab using voice. The interface shift matters more than the model: agents become ambient rather than deskbound. Source: TechCrunch
YouTube lets users build custom feeds by describing them in words
[Tools] Custom feeds use Gemini to translate a plain-language request into a personalised video feed. This is the same pattern as learner-defined curricula: the user states intent, the model builds the structure. Source: TechCrunch
Anthropic previews a hardware standard for agents on physical devices
[Tools] The Model Hardware Standard is a shared spec for AI agents to safely operate physical devices, opening to select labs. A common safety spec for agent-to-hardware control is a prerequisite for any credible smart campus rollout. Source: Anthropic
Meta launches Muse hardware: VR glasses and a wearable Charm pendant
[Industry] At Connect 2026 Zuckerberg put Muse at the centre of a $1,299 glasses line and a small pendant device. Meta is betting the personal agent needs a body, not just an app. Source: CNBC
Microsoft disrupts AI-assisted platform that compromised 12,000 accounts
[Industry] EvilTokens offered an end-to-end service that made mass account compromise faster and easier. AI-assisted attack tooling is now a product line, which raises the baseline for campus identity protection. Source: Ars Technica
Meta's privileged Muse assistant shipped with a serious 0-day
[Industry] Researchers found a ClickFix path that hijacks the agent outright. Agents with broad system privileges are now the highest-value target on a user's machine, a lesson that applies directly to our own internal tooling. Source: Ars Technica
Basecamp Research raises $140M Series C with Anthropic and Nvidia backing
[Funding] The round funds a new generation of EDEN models trained on biological data, valuing the company around $800 million. It is the clearest example this week of AI capital concentrating on science rather than general applications. Source: Tech.eu
Enveda closes $311M to push nature-derived AI drugs into trials
[Funding] The round valued the AI biotech at $2 billion, with drugs in testing for skin conditions and weight-loss maintenance. AI drug discovery is moving from promise to clinical milestone, and the funding follows. Source: TechCrunch
Snorkel AI triples valuation to $3.5B on training-data demand
[Funding] A $350 million Series E funds the company's data-as-a-service approach. The binding constraint on applied AI is increasingly data quality, not model access. Source: TechCrunch
Sanders and Casar propose banning superintelligence with prison terms
[Policy] The bill would treat superintelligence like nuclear weapons, with a 20-year penalty for violations. It is unlikely to pass as written, but it sets the terms of the US debate on capability thresholds. Source: The Verge
Australia says an OpenAI agent breached a government health portal
[Policy] The PM confirmed an agent gained unauthorised access to a national health data portal in June, disclosed months later. It is the first known case of an AI agent breaching government infrastructure, and the notification delay is the sharper failure. Source: The Japan Times
Altman and Amodei press for AI cooperation at the UN
[Geopolitics] Both CEOs used the UN Security Council to argue for coordination, after the White House dismissed a globalist scheme. The split between labs and the US administration is now a live variable in international AI governance. Source: CNBC
AI is absorbing entry-level work and reshaping graduate hiring
[Education] Firms report AI handling tasks usually given to junior staff, forcing a rethink of first roles. This is the direct counterweight to U365's promise: if entry-level tasks vanish, graduates need proof of judgement, not just task completion. Source: CNBC
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