Anthropic weighs a new model as OpenAI gains enterprise ground

Anthropic is reportedly weighing an unnamed new AI model as OpenAI’s GPT-6 Astra gains ground with business customers. The deliberation comes amid pressure to improve profitability and prepare for a possible initial public offering, while CEO Dario Amodei has publicly argued for a slower pace of frontier-AI development.

The model would be a possible response, not a confirmed launch. Reports say Anthropic is evaluating its safety before deciding whether to release it.

Anthropic considers a new model against OpenAI’s advance

The reported deliberations emerged between September 19 and September 21, 2026, after OpenAI released GPT-6 Astra on September 3. OpenAI presented Astra around computer use, software engineering, cybersecurity and professional tasks, while Anthropic’s potential response remains unnamed and under consideration.

That distinction matters for anyone trying to decide whether a new Claude model is imminent: Anthropic has not announced a product name, price, release date or availability for the model described in these reports.

The competitive backdrop is measurable, although it is narrower than a full market-share comparison. Ramp-tracked enterprise AI spending was reported at about 13% for GPT-6 Astra and about 8% for Anthropic’s Claude Fable. OpenRouter also reportedly recorded higher spending on OpenAI models than on Anthropic models in its latest reported week, the first such shift there in more than two and a half years.

The business pressure behind a possible response

For enterprise buyers, the reported spending figures point to a fight over adoption rather than a verdict on every AI product. Ramp measures spending tracked through its corporate-expense platform, while OpenRouter reflects activity on its model-routing platform. Neither figure covers the entire enterprise AI market.

The financial picture is similarly mixed. Anthropic’s annualized revenue run rate was reported above $65 billion by late July 2026, while OpenAI’s was reported above $40 billion in July 2026. Those are annualized run rates, not audited annual revenue, and the figures refer to slightly different observation periods.

A possible IPO adds another layer of pressure. Reports describe Anthropic’s timing as moving toward November or beyond the November U.S. midterm elections, but the company’s offering date and terms remain unsettled. The reported debate is therefore not simply about building a more capable model: it also involves spending on development, profitability and the expectations surrounding a potential public listing.

The tension between speed and safety

On September 12, Dario Amodei published an essay calling for a slower pace of frontier-AI capability development. His position was that safety measures and governance need time to keep pace with increasingly capable systems.

The later reports about a possible competitive model put that position under pressure. Anthropic may be considering a faster commercial response to OpenAI while also evaluating whether the prospective system meets its safety requirements. Those are separate decisions: competitive urgency does not amount to a release commitment, and a safety review does not establish a final safety result.

The company’s public safety posture has also included discussion of external scrutiny. Anthropic and Accenture’s Faculty division were reported on September 21 to be preparing independent assessments of frontier models and safeguards. The proposed work is intended to give external evaluators access comparable to that of internal employees, with narrow limits for security-sensitive, legally privileged, commercially sensitive or third-party confidential information.

What a possible Anthropic IPO changes

A public listing would give investors a direct role in judging the balance between model investment and financial returns, but the timing remains prospective. The reported possibilities range from November to a later date, rather than a settled offering schedule.

That context helps explain why enterprise adoption appears so important in the model discussion. A stronger position among business customers could support Anthropic’s commercial case, while the cost of training and evaluating frontier models increases the pressure to show durable returns. The reported revenue run rate provides one measure of scale, but it does not settle profitability or establish the terms of a future offering.

Anthropic and Accenture prepare external evaluations

The Faculty development is a separate current event from the unnamed-model deliberation. Anthropic and Accenture were each expected to invest at least $1 billion over five years in evaluation capacity, according to the reported plan. Anthropic was also said to be in discussions with Model Evaluation & Threat Research about embedded evaluation.

The proposed evaluators would be able to publish important findings, subject to narrow redactions. That arrangement would give outside scrutiny a formal role in Anthropic’s safety process while the company weighs how to respond to OpenAI’s enterprise momentum.