Anthropic Weighs New Claude Model as OpenAI's Astra Surges

Anthropic is weighing the release of another artificial-intelligence model as competitive pressure from OpenAI's recently launched GPT-6 Astra intensifies, according to a Reuters report published on September 19 that cites three people familiar with the matter. The company behind the Claude model family is evaluating the safety of the prospective model and has not publicly announced its capabilities, launch date or commercial terms, so the potential release remains an internal strategic option rather than a confirmed launch.

The deliberations come only weeks after Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 on September 1. Anthropic describes the two products as the same underlying model with different safeguard configurations: Fable 5.1 is generally available, while Mythos 5.1 is restricted to vetted organisations working in areas such as cybersecurity and life sciences. The company said Fable 5.1 should cost approximately 25 percent less than Fable 5 for typical token-billed workloads because of lower cache-read pricing.

Why Astra's early traction is forcing the issue

OpenAI launched GPT-6 Astra on September 3, describing it as its most capable model for demanding work spanning coding, computer use, research and professional tasks. According to Reuters, Astra has quickly gained ground with business users. Data from Ramp, a corporate-card and expense platform, shows Astra accounting for around 13 percent of tracked enterprise AI spending, compared with approximately 8 percent for Anthropic's Claude Fable. OpenRouter, a model-routing marketplace, saw spending tilt toward OpenAI during the latest reported week for the first time in about two and a half years. Those indicators cover particular customer populations rather than the entire AI market, but they show Astra building early commercial momentum.

Frontier AI product cycles are getting shorter because developers compete at once on capability, cost and enterprise adoption. A model that leads performance benchmarks for only a few months can lose commercial momentum quickly when a rival ships stronger coding, reasoning or agentic features. For Anthropic, shipping a stronger model quickly could preserve customer momentum, but moving too fast could raise development spending, add safety-testing requirements and shorten the commercial life of products launched only weeks ago.

A safety-first brand under commercial strain

This is a delicate spot for Anthropic because the company has built its market position partly around model safety and reliability. Chief executive Dario Amodei recently published a 3,800-word open letter calling on development labs to deliberately slow the pace of leading-edge model upgrades so researchers have time to map and defuse emergent hazards. That plea won backing from OpenAI's Sam Altman and xAI's Elon Musk, as reported by Bloomberg. A rapid follow-up release would have to meet Anthropic's own deployment standards while delivering enough extra value to justify another product transition, and it would sit uneasily beside that public call for restraint.

Reuters reported that Anthropic is evaluating the safety of the prospective model as part of its deliberations. The company's current product structure already separates frontier capabilities from higher-risk access: Mythos 5.1 is available only to vetted organisations under a different safeguard structure, while Fable 5.1 offers the same underlying model under restrictions designed for broader use. The company also publishes system cards documenting safety evaluations and deployment decisions across its major Claude generations, adding an evidence step before any commercial release.

Revenue is soaring, and an IPO is on the horizon

Reuters reported that Anthropic's annualised revenue run rate exceeded 65 billion dollars by the end of July, up from roughly 9 billion dollars at the end of 2025. The company is projecting annual revenue of approximately 190 to 200 billion dollars by 2028, according to people familiar with its financial plans. OpenAI's annualised revenue run rate was reported above 40 billion dollars in July. As reported by CryptoBriefing, which relayed the Reuters reporting, the idea behind a new Claude model release centres on using a fresh product launch to support a stronger valuation heading into an initial public offering.

Anthropic confidentially filed for a US IPO in June, and Reuters previously reported that a pre-IPO revolving credit facility under discussion was expected to exceed 10 billion dollars. On September 19, Reuters added that the company could delay a listing until after the November 2026 US midterm elections. Anthropic declined to comment on the reported IPO timing. A listing would expose the economics of model development to far greater scrutiny, with investors asking how quickly revenue can grow relative to computing expense and the capital required to secure increasingly large amounts of AI infrastructure.

Customer concentration may matter as much as model performance. Reuters reported that Meta Platforms is one of Anthropic's largest customers but is seeking to reduce its reliance on external models as it builds more internal AI capability; Meta did not comment on that account. Large customers can accelerate revenue fast, but they also create concentration risk if they eventually move workloads in-house, negotiate lower prices or spread spending among several suppliers.

What happens next comes down to three open questions. The first is whether the model under evaluation reaches public release at all; Anthropic has not announced it, so any product name, timing or capability claim beyond Reuters' reporting would be premature. The second is enterprise spending data in the coming weeks, which will show whether Astra's momentum holds or whether Anthropic answers successfully with Fable 5.1 and whatever follows. The third is the IPO, where public investors would take on exposure to one of the fastest-moving and most capital-intensive competitive cycles in modern technology.

For more on the Reuters reporting, see Finimize's breakdown of the story.