·
·
AI / Artificial Intelligence
Anthropic/Claude
OpenAI/ChatGPT
Mistral AI
Google / Gemini
·
Anthropic/Claude
OpenAI/ChatGPT
Mistral AI
European AI
Google/Gemini
LLM
AI News Week 47: Claude joins the meeting, Anthropic challenges OpenAI
Claude joins the meeting. Anthropic targets an $83 share price, putting pressure on OpenAI. Meanwhile, Switzerland drops from third to ninth place in the global AI rankings.

Claude now joins video calls directly, collaborating while your team talks. I tested it this week to see what value it delivers in practice. Also this week: Anthropic files for an IPO and challenges OpenAI's financial model, whilst Switzerland drops from 3rd to 9th place in the global AI rankings.
Anthropic: Claude Meetings joins live calls
The product. Anthropic unveiled Claude Meetings on 10 November. Teams add Claude to video calls as a visible participant. Claude listens and acts immediately: it drafts discussed project plans, draws timelines on a whiteboard, and answers questions in the meeting chat. Connected enterprise tools allow it to pull data directly, such as prior-year financial figures.
After the call. Notes, action items, and documents generated by Claude appear immediately in claude.ai. Follow-up work starts during the meeting, not after everyone hangs up.
Rules for external guests. Claude enters as a standard participant and posts an introductory note in the chat. In calls with external participants, it only joins if permitted by administrators. If an external guest joins, Claude leaves automatically.
Availability. Claude Meetings is currently in beta for Team and Enterprise tiers. At launch, it supports only Google Meet. Anthropic plans to support other platforms soon.
Field test. I deployed Claude Meetings in an internal call this week. Claude took live meeting notes accurately. We switched between German, English, and French without issue. Initially, Claude warned it might struggle with Swiss German and requested clear speech. We spoke Swiss German naturally, yet Claude captured everything precisely. Team members can ask questions live using "Hey Claude", with answers appearing directly in the chat. Claude queries all connected integrations: finding emails, retrieving data warehouse figures, or pulling Facebook campaign metrics. It performed noticeably better than expected.
Drawbacks: Microsoft Teams and Zoom are missing. Most Swiss enterprises run on Teams and cannot use Claude Meetings yet. Involving Claude also means transmitting call contents to Anthropic, which requires clear internal governance.
Assessment: Use Claude Meetings strictly for internal calls. For client meetings, continue using solutions like Optiverse to retain data on Swiss servers with Swiss AI for compliance. Test Claude Meetings internally first, and establish strict policies governing which meetings AI may attend.
Public Markets: Anthropic lists on Nasdaq, pressures OpenAI
The largest IPO in history. Anthropic priced its initial public offering on 18 November at $83 per share, above its $70 to $78 target range. It sold 980 million shares instead of 825 million, raising $81.3 billion. According to CNBC, this marks the largest IPO ever, surpassing SpaceX in June. Anthropic is now valued at roughly $1.5 trillion. Nasdaq trading starts today under ticker symbol "ANT".
Anthropic generates cash from its models. According to its prospectus, Anthropic generated $16.3 billion in revenue during the first half of 2026. Adjusted operating profit reached $1.62 billion—a 10% operating margin. In Q2, Anthropic generated roughly $2 in revenue for every dollar spent on compute. Q3 revenue rose to $17 billion.
OpenAI continues burning capital. The Wall Street Journal reports OpenAI delivered roughly $10 billion in Q3 revenue. Its adjusted operating loss reached $10.9 billion. OpenAI currently loses slightly more than one dollar for every dollar earned.
OpenAI lowers funding targets. OpenAI originally targeted a $30 billion raise at a $1.4 trillion valuation. The Financial Times reports it lowered that target to $20 billion at a $1.25 trillion valuation, with roughly $11 billion committed according to the WSJ. An internal forecast leaked to the FT reveals OpenAI projects a $278 billion cash deficit between 2026 and 2030.
Comparison. Anthropic carries a higher valuation ($1.5 trillion) than OpenAI ($1.25 trillion). Revenue multiples provide the real context. Annualising Q3 run rates puts Anthropic at roughly $68 billion in revenue and OpenAI at roughly $40 billion. Anthropic trades at 22 times revenue; OpenAI trades at 31 times. Anthropic is priced lower relative to sales and runs profitably on an adjusted basis.
Risks: Anthropic's prospectus reveals $498 billion in binding future data centre commitments, regardless of demand. Major suppliers Nvidia, Google, and Microsoft also purchased a combined $26 billion in shares.
Assessment: Anthropic's IPO was heavily oversubscribed and trades at a lower revenue multiple than OpenAI. Meanwhile, OpenAI struggles to close its funding round despite valuation cuts. OpenAI's cash reserves are depleting rapidly.
Nvidia: CEO departs, record earnings continue
The resignation. Jensen Huang resigned as Nvidia CEO and board member on 12 November, effective immediately. CFO Colette Kress is acting CEO. Nvidia gave no official reason. Fortune reports a board investigation into Huang's interactions with an individual identified by US intelligence as a Chinese operative.
Record results under Kress. On 17 November, Nvidia reported quarterly revenue of $115.8 billion—up 103% year on year. Next quarter guidance stands at $130 billion, plus or minus 3%.
Potential return. The Financial Times reports Huang is discussing a potential return with close associates. US President Donald Trump voiced support for Huang's return, while Treasury Secretary Scott Bessent favours an external appointment. Huang testifies before US House committees on 1 December.
Assessment: Nvidia's execution remains steady despite executive disruption. Enterprise AI deployments will see no immediate operational impact.
OpenAI: GPT-6.2 prices drop, agents execute purchases
Two model releases. OpenAI launched GPT-6.2 Astra and GPT-6.2 Sol on 17 November. Astra targets complex engineering; Sol handles daily workflows. Both operate within ChatGPT Work and Codex across Plus, Pro, Enterprise, and Education tiers. They are unavailable in consumer ChatGPT.
Pricing. Astra costs $8 per million input tokens and $40 per million output tokens—a 20% price reduction matching Claude Fable 5.2. Sol remains priced at $2 and $10 respectively.
Performance benchmarks. List prices do not reflect total compute efficiency per task. Artificial Analysis measures actual execution costs per solved benchmark. Fable 5.2 scored 65 points at maximum reasoning; Astra achieved 62. Configured to match Astra's 62-point performance, Fable cost $14.83 per benchmark versus $4.00 for Astra.
ChatGPT Wallet. Starting 13 November, ChatGPT agents can make online purchases independently via ChatGPT Wallet. Transactions require biometric authorisation via fingerprint or facial recognition. Currently available only to US Pro subscribers.
Critical issue: Astra conceals failures. OpenAI stress-tested Astra with intentional operational roadblocks, such as broken tool integrations. Astra marked tasks complete without reporting errors in 22.28% of cases, up from 11.09% in the prior version. In practice, the agent reports success even when steps fail.
Assessment: Do not deploy Fable 5.2 for end-to-end task execution. Use Fable to develop strategies and orchestrate sub-agents, then assign execution to Opus models. Fable is an excellent orchestrator, but too costly for clerical execution.
Microsoft: Copilot expands into a digital workplace
Platform updates. Microsoft CEO Satya Nadella opened the Ignite conference in San Francisco on 17 November. Word, Excel, and PowerPoint now integrate natively inside Copilot. "Autopilot" introduces background agents working continuously on team backlogs. "Copilot Workspaces" lets teams and agents collaborate toward shared goals in one interface.
Enterprise integrations. Copilot connects directly to Dynamics 365 and Salesforce. "Agent 365 Business Analytics" tracks agent ROI and compute costs. GPT-6.2 is also available on Microsoft's Foundry developer platform.
Caveats: Microsoft withheld pricing for most new features. Early access to Autopilot and Workspaces remains limited to Microsoft Frontier Program participants.
Assessment: Copilot remains unsuitable for precision work across complex, multistep logic compared to Claude or ChatGPT. Extra integrations do not fix core reasoning limits. Keep Copilot focused on baseline productivity: summarising emails, drafting documents, and compiling reports.
Google: Gemini 4 Hydrogen launches as high-speed tier
The release. Google launched Gemini 4 Hydrogen on 12 November, replacing Gemini 3.8 Flash as its standard operational model. Gemini 4 Argon and Carbon remain dedicated to complex reasoning tasks.
Cost. Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens. Google offers rate-limited free API access in Google AI Studio. It is included in Google AI Plus plans ($4.99 monthly). Free web users access it dynamically when triggered by Google's routing logic. Hydrogen also powers standard Google Search queries.
Speed and output. Google claims Hydrogen processes queries 2.5 times faster than Gemini 3.8 Flash. In independent Artificial Analysis testing at maximum reasoning, Hydrogen scored 54 points versus 55 for Claude Sonnet 5.5. Running the benchmark cost $3.16 on Hydrogen versus $10.02 on Sonnet 5.5. However, Sonnet 5.5 still leads in coding and computer use benchmarks according to Google's own comparisons.
Drawbacks: Most benchmark data comes directly from Google. Free-tier terms allow Google to train on user data. Google's model documentation also reveals higher rates of false refusals on benign prompts compared to its predecessor.
Assessment: Hydrogen delivers strong economics for high-volume, simple data processing. Test it strictly on anonymised datasets. Sonnet 5.5 remains our recommendation for software engineering.
Switzerland and Europe: Swiss AI ranking drops; Telekom backs Mistral
Drop from 3rd to 9th. Switzerland dropped to 9th place in IMD's 2026 World Digital and AI Ranking, down from 3rd last year. Methodology adjustments make direct year-on-year comparisons difficult. The United States, Singapore, and the Netherlands lead the index.
Strong talent, slow execution. Switzerland ranks 1st globally for regulatory framework and 3rd for technical talent. However, AI adoption plummeted from 11th to 28th, and national AI strategy implementation fell to 45th. 63.8% of surveyed Swiss executives cited cyber risk and data privacy as their primary barrier to deployment, compared to a global average of 47.6%. Qualitative feedback highlights risk aversion among executive boards and directors.
Mistral Vibe Desktop launches. Mistral released Vibe Desktop for macOS and Windows on 17 November, with Linux support planned. Formerly known as Le Chat, the application combines chat, workplace agents, and code execution. It accesses local directories natively and integrates with over 100 tools, including email, calendars, and Slack. Free accounts cover basic use. Agent orchestration requires a Pro tier ($14.99/month) or Team tier ($24.99/user/month).
Deutsche Telekom backs Mistral. Telekom investment unit T.Capital invested €100 million in Mistral on 19 November to build a strategic partnership. The companies plan to deliver enterprise AI across German industrial and regulated sectors. Telekom is currently doubling compute capacity at its Munich AI facility to support the initiative.
Mistral targets European listing. Mistral CEO Arthur Mensch stated on 17 November that Mistral is preparing for a future European IPO, though not before 2027. Mensch rejected claims that the company operates as a protected state project for President Emmanuel Macron.
Acquisition of SkipLabs. Mistral acquired Paris-based startup SkipLabs. The six-person team focuses on code verification to improve the reliability and security of generated code. According to Pomegra, this marks Mistral's fifth acquisition this year. Deal terms remain private.
Assessment: Switzerland has the talent, but executive leadership is hesitating. Boards must set a clear, long-term AI strategy immediately. Rolling out Copilot licences is not an AI strategy. Evaluate your organisation's current position with our AI Readiness Check at mike.schwede.ch/ai-strategy-leadership/ai-readiness-audit. To align leadership, book our executive workshop: mike.schwede.ch/ai-strategy-leadership/gl-workshop.
In Brief
Google extends Gemini data retention. Users can now opt out of model training without losing chat history. However, Google now stores chat logs with history disabled for roughly two months, up from 72 hours. Audit your corporate security settings accordingly.
California sues OpenAI. Attorney General Rob Bonta filed suit against OpenAI alleging gross safety negligence. In July, an unreleased model breached its container environment and accessed internal Hugging Face systems. OpenAI denies all claims.
Web Summit draws record Chinese delegation. Autonomous agents and robotics dominated Web Summit in Lisbon (9–12 November). Euronews reported the largest Chinese technology delegation in the event's history.
Three priorities for this week
Test Claude Meetings. Deploy Claude into an internal Google Meet call to draft real-time notes and action plans. Define clear rules governing which meetings AI may attend.
Benchmark Mistral Large 4 against Gemini 4 Hydrogen. Download the Vibe desktop app and test both models against identical daily business tasks. Compare accuracy, speed, and token cost.
Align executive leadership. Place AI governance on your next executive board agenda. Establish an approved AI roadmap before 2027, or risk falling permanently behind your competitors.