Anthropic puts part of its equity into a new Public Benefit Fund after a $65 billion round

On October 18, Anthropic announced the creation of the Anthropic Public Benefit Fund. It is a new structure to which the company is transferring part of its shares. The mechanism is designed so that the fund's stake increases as Anthropic itself grows. The announcement names neither the size of the initial stake nor the rules for managing the fund's assets, so it is too early to judge the scale. The timing is what makes this notable: in May the company closed a $65 billion Series H round at a $965 billion valuation, and on June 1 it confidentially filed a draft S-1 with the SEC.

Why the fund is appearing now

A company preparing for a public offering usually fixes its ownership structure in advance. A charitable fund with a growing stake solves two problems at once. It gives investors a predictable picture of the cap table, and it shows regulators and the public that part of the company's value is set aside for public purposes. The fund adds to the existing Long-Term Benefit Trust. In July, former Fed chair Ben Bernanke joined that trust, and in August Anthropic hired Mariano-Florentino Cuéllar in a new role as Chief Global Affairs Officer. Taken together, it looks like consistent preparation for life as a public company.

We do not know the details of the structure. They will have to be found in the S-1 once the document becomes public.

An autumn on the assembly line: four models in a month

At the same time, Anthropic has refreshed almost its entire lineup. Here is what the announcements say:

• Claude Mythos and Fable 5.1 (September 15): positioned as the most advanced models for coding and knowledge work, with notable research capabilities.
• Claude Opus 5.5 (September 22): the leader in agentic coding and knowledge work, and 40% cheaper than Opus 5 on typical workloads.
• Claude Sonnet 5.5 (September 28): runs 30%+ faster than Sonnet 5 and costs up to 30% less for most tasks.
• Claude Haiku 5.5 (October 7): the fastest small model, built for summarization, subagents and browser work.

The trend is clear: the emphasis is on cost of execution and speed, not only on quality gains. For agentic systems, where a single task spawns dozens of calls, that matters more than polished benchmark numbers. Pairing an expensive planner with cheap Haiku subagents is becoming an economically sensible setup.

People, regions and access to sensitive capabilities

The commercial expansion is moving on a broad front. Barclays is extending its partnership and rolling out Claude across its global operations. The company opened an office in Singapore and signed a memorandum with the country's government; offices in Seoul and Milan (its sixth in Europe) opened earlier. A separate bet goes to talent: $100 million is going into the Claude Frontier Academy, with the goal of training 10,000 Frontier Deployed Engineers by the end of 2027. The first cohorts include specialists from Accenture, Bain, CBA, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk.

A separate track is controlled access. The expanded Cyber Verification Program gives vetted security professionals advanced cyber capabilities and relaxed blocking classifiers. The Life Sciences Verification Program opens the Mythos, Opus and Sonnet models with looser restrictions for biological tasks. This is a model in which the strictness of safeguards depends on the user's verified role. Alongside it sits the Model Hardware Standard: a research preview of a specification under which agents could safely control physical devices, for now aimed at scientific laboratories and advanced manufacturing.

Grants go to the science and research side: 10,000 scientists get Claude for free at launch, $5 million goes to assessing AI's impact on user wellbeing, $10 million to Canadian research, and up to $50,000 in credits to rare-disease projects.

What teams should check

Developers should recalculate their budgets: the claimed 30–40% savings on Opus and Sonnet 5.5 need to be verified on your own traces, not on demos. Teams with agentic pipelines should test Haiku 5.5 as a subagent and on browser steps. Security specialists and biologists should find out whether they meet the criteria of the verification programs. Investors and market analysts should watch for the S-1 filing: that is where the actual parameters of the fund, the trust and the share of the company being set aside for public purposes will appear.