AI Brief.

The week in 30 seconds

AI usage costs can halve in one release, so continually evaluate the models available in your enterprise. A Big Four firm runs outsourced finance on one platform; Microsoft's buyout offer reportedly reaches its AI teams too.

Top stories

01   Business   Anthropic · IPO filing

Anthropic's IPO filing shows most of its revenue is billed by use

What changed: Anthropic has filed for a Nasdaq listing, and four-fifths of its revenue is billed on actual usage rather than subscriptions. Existing customers spend nine times what they did a year ago. An amended filing on 2 November shows nearly $17bn of third-quarter revenue, against operating costs of around $17.5bn, according to reports on the filing. Anthropic has committed $895bn to computing power and data centre leases.

Why it matters: A supplier with fixed compute bills and income that rises with use needs its customers to keep using more.

02   Professional services   Tax and accounting · PwC UK and FloQast

PwC UK moves its outsourced finance work onto accounting platform FloQast

What changed: PwC UK will now use FloQast directly to run its finance managed services. Its teams will do clients' month-end closing, account reconciliations and regulatory compliance in the platform, which a PwC UK partner calls 'AI-enabled delivery'.

The catch: It's unclear if PwC firms in other countries will run client work this way.

Why it matters: A Big Four firm is moving from advising on these platforms to running client work on one.

03   Work and jobs   Microsoft · Voluntary buyouts

Microsoft offers buyouts to up to 60,000 staff, reportedly including AI teams

What changed: Microsoft is offering voluntary buyouts to up to 60,000 employees in the US, UK, Canada and India, about a quarter of its 223,000 or so staff. Sources told The Verge that teams working on Copilot, CoreAI and other AI products are eligible.

Why it matters: The buyouts aim to trim staff without another round of layoffs, and The Verge's sources say even AI teams are in scope.

04   Models and tools   OpenAI · Pricing

OpenAI's new mid-range model halves its API prices

What changed: OpenAI has released GPT-6.1 Terra at half the API price of its predecessor. It costs $1 per million input tokens (units of text) and $5 per million output tokens, against $2 and $10 for Anthropic's Claude Sonnet 5.5 and $2 and $12 for GPT-5.6 Terra. Artificial Analysis scored it 48 on its Intelligence Index at launch, between OpenAI's larger GPT-6.1 Sol (52) and its predecessor (42).

Models and tools

What's new this week

Research worth knowing

Research   OpenAI · Maths and computer science

OpenAI says its models have disproved a widely believed assumption in computer science

OpenAI says its models have disproved the Exponential Time Hypothesis. Three outside computer scientists, two of whom helped propose the hypothesis in 1999, will examine OpenAI's proof against it.

The catch: These are OpenAI's claims until the checks report.

Research   Anthropic study

Experts adopt an AI agent's coin-flip views as their own

What changed: Anthropic gave 50 professional analysts an AI agent whose lean on each judgement call was secretly set by a coin flip. On four analytical choices they went with its lean 86% of the time, and 77% later explained their choices with its arguments, but only 15% credited the agent.

The catch: Anthropic tested its own model with a lean strengthened on purpose, and says the effect might be smaller without that.

Try this

Make your AI argue both sides first

From the study above: before your firm's approved AI tool gives its view on a judgement call, ask it to argue both sides.

Also this week