Anthropic Sets the Stage for IPO, Google Launches Carbon, Nvidia Investigates

Google introduces Gemini 4 Carbon with restricted access and announced pricing. Anthropic clarifies its IPO range, updates its figures, and adds a warning about recursive self-improvement. Meanwhile, Nvidia announces an independent investigation into its CEO, and AI groups are strengthening their anti-espionage training.

Google restricts access to Carbon and announces pricing

Google has launched Gemini 4 Carbon, initially restricting access to a limited number of cybersecurity defense companies. The model is priced at $2 for every million input tokens and $10 for every million output tokens, a pricing structure that Google claims is equivalent to that of Argon. According to the company, significant progress has been observed in coding, agentic tasks, and cybersecurity, with Carbon ranking first in the Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 and GPT-6 Astra. Google also describes long-duration usage, with internal teams having seen it operate for days on complex engineering tasks.

Anthropic releases Mythos 5.2 and Fable 5.2 and lowers a price

Anthropic has unveiled the Mythos 5.2 and Fable 5.2 models. The price of Fable has been reduced from $10 to $8 per million input tokens. The company reports that this version of Fable provides responses 30% faster.

Anthropic IPO: range, timeline, and targeted net proceeds

For its IPO, Anthropic has set a price range of $70 to $78 per share, which could correspond to a valuation of $1.404 trillion. The company plans to begin trading around November 19 on the Nasdaq under the symbol ANT, with up to 20% of the offering allocated to retail investors, and anticipates net proceeds between $84.6 billion and $93.7 billion. Its updated filing mentions an annual revenue of $81 billion by the end of October, an increase of about $10 billion from September, as well as a net loss of $104 billion over nine months, primarily due to non-cash charges related to convertible notes.

Scaling risks and governance at Anthropic and Nvidia

Anthropic has added a risk factor in its revised S-1/A indicating that it may be nearing the threshold it defines for recursive self-improvement, reporting that internal estimates suggest an imminent approach to the definition, as per its Responsible Scaling Policy, of significantly accelerated progress attributable to AI. Meanwhile, Nvidia has announced the opening of an independent investigation concerning its CEO Jensen Huang, following reports of a Chinese espionage operation targeting him; the company released a message from Jensen Huang acknowledging an affair. Nvidia's board states that it has no reason to believe that national security or company secrets have been compromised. A memo sent to U.S. decision-makers mentions the recruitment of Alice Huang, former Megaspeed, by the Chinese Ministry of State Security to approach the executive.

Strengthened counter-espionage and shadows over buyers in China

OpenAI, Anthropic, and Google have implemented counter-espionage training to raise awareness among their teams about risks such as bribery, eavesdropping in cafes, and attempts at seduction by foreign agents, according to U.S. business press. In China, according to the New York Times, doubts surround New Matrix, an approved beneficiary of H200 chips: the company reportedly occupies a vacant office without a listing in the building directory, and its owners are linked to a Chinese private equity firm that has been blacklisted in the U.S. since 2024.