Astra leads enterprise spend: 13 percent to Claude Fable's eight, per Ramp data

Astra leads enterprise spend: 13 percent to Claude Fable's eight, per Ramp data

About two weeks after OpenAI launched GPT-6 Astra on 3 September, Anthropic is weighing the release of a new model to counter OpenAI's momentum — even as the company is reportedly considering postponing its IPO until after November's midterm elections, according to a Reuters exclusive built on anonymous sources.

Anthropic is thus considering rolling out a new AI model in response to OpenAI's GPT-6 Astra, according to three sources familiar with the matter whom Reuters has spoken with. The report, by Echo Wang, Krystal Hu and Milana Vinn and published on 18 September, places the deliberation in a tight three-part context: Astra launched on 3 September, CEO Dario Amodei urged the industry on 12 September to slow the pace of capability development, and the company's long-awaited IPO has already been delayed once and could now be pushed until after the election. Anthropic declined to comment (Reuters via MSN).

How it's explained

The report does not say which model is at issue, when it might arrive, or what it is supposed to be able to do. The only concrete content of the deliberation is that Anthropic, according to one of the sources, is evaluating the safety of the next model as part of the release decision. That places the prospective launch inside the company's own release process, not purely as a competitive reaction — but the sources describe OpenAI's momentum as the trigger.

It is worth underlining what kind of story this is: everything rests on three anonymous sources. There is neither official confirmation, nor named individuals, nor dates. Anthropic has not commented. Everything else in this story — including the IPO question — is live and unsettled.

The numbers behind the Astra momentum

What makes the story timely are the measurements showing that Astra has actually moved something:

• Ramp: GPT-6 Astra accounts for roughly 13 percent of enterprise AI spending tracked by the expense platform Ramp, versus roughly 8 percent for Anthropic's Claude Fable, according to recent data cited by Reuters.
• OpenRouter: Users of OpenRouter spent more money on OpenAI models than on Anthropic models last week — the first time in more than two and a half years that OpenAI has led on that measure.

At the same time, Anthropic is far from a weak player in financial terms. The company's annualized revenue run rate passed $65 billion at the end of July, up from around $9 billion at the end of 2025. OpenAI passed $40 billion in July. It is a picture of two companies growing quickly, but with Anthropic still leading on revenue — while losing position on usage and spending measures.

The Meta signal

Another detail weighing on the picture: Meta Platforms has been among Anthropic's largest customers, but is now considering reducing its use of Anthropic's models as the company builds more AI capability in-house, according to people familiar with the matter. Meta did not respond to Reuters' request for comment. This too is anonymously sourced material — but if accurate, it is a concrete loss for Anthropic on the customer side, not just a measurement.

Amodei's own appeal

This is the backdrop that makes the story harder than an ordinary competitor story: Six days before the Reuters report, on 12 September, Amodei published a 3,800-word essay urging the industry to slow the pace of capability development. "We must slow the pace at which we improve the capabilities of AI models," Amodei wrote, according to Reuters (same source).

It is tempting to read this as a contradiction. But it is worth holding on to what has actually been reported: Anthropic is weighing a new model, not a decided launch, and safety evaluation is part of the deliberation according to one source. AIMag draws no conclusion about Amodei's motivation — we report the tension as it appears in the sources.

The IPO on hold

The third thread in the story is the IPO. According to two sources familiar with the matter, Anthropic could push the IPO until after November's midterm election — and the election is not expected to have a major impact on the offering itself. The IPO has already been delayed from earlier plans. Reuters has previously reported that marketing is expected to begin no earlier than mid-October.

That means the listing, like the model release, is still an open question — not a decision.

Investors disagree

A fourth layer of the report that is often overlooked: The investor picture is split. Some investors have told Reuters that they see Astra as a threat to Anthropic's position, while others believe the threat is exaggerated given Anthropic's lead on the enterprise side and strong revenue run rate. There is thus no consensus that Anthropic is actually losing ground at the core — only concrete measurements pointing in that direction, met by a revenue trajectory pointing the opposite way.

What we don't know

• Whether Anthropic will actually launch a new model, and what it might be called or be able to do.
• Whether the IPO is actually postponed until after the election.
• How much weight the safety evaluation actually carries in the decision-making process.
• Whether Meta's reduced usage will actually happen, and to what extent.
• Whether the Astra momentum persists, or whether the Ramp and OpenRouter measurements are a short-lived launch bump.

Why it matters

Even with all the caveats about anonymous sources, this is a story about something larger than one company: competitive pressure from a concrete, newly launched model can push players to act faster than their own stated preference for a slower pace. If that happens at Anthropic specifically — the company built on its safety profile — it says something about how little room the industry actually has to follow its own appeals. But that is a conclusion we cannot draw yet. For now, what stands firm are the numbers, the dates, and the unresolved questions.