Anthropic Eyes New Model to Counter GPT-6 Astra's Rise
On September 19, 2026, Reuters published an exclusive report: Anthropic is weighing the release of a new AI model to counter commercial momentum from OpenAI’s GPT-6 Astra, according to three sources familiar with the deliberations. The potential launch would come ahead of the company’s expected IPO and in direct tension with a public call from its own chief executive for an industry-wide slowdown.
For enterprise AI buyers and GTM leaders, the story is not just about which model ships next. It is about how the two dominant enterprise AI vendors are navigating an unprecedented collision of competitive pressure, safety commitments, and capital markets.
The Numbers Driving the Decision
The proximate cause of Anthropic’s internal debate is a single data point: enterprise spending has begun shifting.
According to Ramp, the corporate expense platform, OpenAI’s GPT-6 Astra now accounts for about 13% of enterprise AI spending tracked on Ramp. Anthropic’s Claude Fable accounts for about 8%. Astra launched on September 3, 2026, barely two weeks before the Reuters report, touting gains across computer use, software engineering, cybersecurity, and professional work.
The gap is not a collapse of Anthropic’s position. Claude Fable still leads the broader market. Anthropic’s annualized revenue run rate topped $65 billion by the end of July 2026, up from roughly $9 billion at the end of 2025. That is a sevenfold increase in revenue in roughly seven months. OpenAI’s annualized run rate passed $40 billion in July.
But for investors preparing to evaluate an IPO, a 5-point spending share deficit to a model that launched two weeks ago is a data point they cannot ignore.
| Metric | Anthropic | OpenAI |
|---|---|---|
| Annualized Revenue (July 2026) | ~$65B | ~$40B |
| Enterprise AI Spend Share (Ramp) | 8% (Claude Fable) | 13% (GPT-6 Astra) |
| Model Referenced | Claude Fable 5.1 | GPT-6 Astra |
| IPO Plans | Potentially post-Nov midterms | Not in 2026 (per Altman) |
| CEO Pacing Stance | Called for slowdown (Sept 12) | Endorsed pacing framework |
The Contradiction at the Heart of the Deliberation
The timing creates an uncomfortable public contradiction for Anthropic.
On September 12, Dario Amodei published a 3,800-word essay, “We Must Pace the Frontier,” calling on the global AI community to slow the release of new capabilities to address safety concerns. “We must slow the pace at which we improve the capabilities of AI models,” he wrote. The essay proposed three structural reforms: coordinated safety standards across frontier labs, independent evaluators with employee-level access inside companies, and global governmental coordination.
The essay drew support from Sam Altman, Elon Musk, Satya Nadella, and Demis Hassabis. Anthropic announced a concrete step toward implementation just days later, embedding independent evaluators from Accenture’s Faculty unit with employee-level access inside the company.
And now, per three Reuters sources, Anthropic is weighing a new model release anyway.
Anthropic declined to comment to Reuters. One person familiar with the deliberations said the company is still evaluating the safety of its candidate model as part of the release decision. Some internal discussions involve how to balance model investment against profitability timelines as rising interest rates focus investors on when profits arrive, not when models ship.
The framing Anthropic will likely use, if it releases a model, is consistent with Amodei’s essay: pacing is about coordinated standards and evaluator access, not a unilateral halt. A company can call for industrywide coordination while continuing to develop and release models that have passed its own safety evaluations. That argument is defensible. It will still attract scrutiny.
What the IPO Context Adds
The Reuters scoop lands at a specific moment in Anthropic’s capital markets calendar. The company had originally targeted an October IPO. That has slipped. Two people familiar with the matter told Reuters that Anthropic could push the offering to after November’s US midterm elections, with IPO marketing expected to begin in mid-October at the earliest.
For potential IPO investors, the question is not simply “is this a good model company.” It is “how durable is the revenue lead against OpenAI, and what is the risk that Astra accelerates market share transfer.”
OpenAI has removed some competitive pressure on the calendar. Altman confirmed on Saturday that OpenAI will not go public in 2026, citing AI safety concerns. That gives Anthropic a cleaner public markets window. But it also frames the race differently: Anthropic is approaching public markets while its main competitor is using safety rhetoric to delay its own listing, and both companies are now managing the tension between capability releases and the shared narrative around responsible development.
What Enterprise Buyers Should Take From This
For enterprise AI leaders, the Anthropic-vs-Astra dynamic has three practical implications.
Spending data is a lagging indicator. The Ramp data shows Astra at 13% versus Claude Fable at 8%, but this reflects less than three weeks of post-launch enterprise adoption. Enterprise procurement cycles move slowly. A company that signed a Claude enterprise agreement six months ago is not switching models because Astra is good. The more meaningful signal will be what new enterprise workloads are landed on which model over Q4.
Anthropic’s revenue lead is real. A $65 billion annualized run rate against OpenAI’s $40 billion means Anthropic has significantly more embedded enterprise business. That is not at risk from a single model launch. Enterprise AI adoption is not a winner-take-all sprint, and switching costs inside large companies are substantial, as multiple investors told Reuters.
A new Anthropic model is material for your planning. If Anthropic releases a successor to Claude Fable 5.1 before the end of the year, that changes the evaluation landscape for enterprises currently mid-procurement on AI platform decisions. Teams doing model selection right now should build evaluation criteria that can accommodate a late-2026 Anthropic release, not just compare the current frontier lineup.
Enterprise procurement teams would be well-served to model three scenarios: Anthropic ships a new model before IPO, Anthropic ships after IPO, and Anthropic holds the release into early 2027. In each case, the contract and vendor commitment terms differ. Enera helps enterprise teams build model-agnostic deployment architectures so that a competitive model release becomes an advantage, not a disruption.
The Bigger Picture
This story is one episode in a longer arc. Earlier analysis on Enera covered how Amodei’s pacing essay created new governance expectations for enterprise AI teams. The Reuters scoop adds a new layer: the companies calling for coordinated slowdowns are simultaneously under pressure to ship, because their market positions depend on it.
That tension will not resolve with a single model launch. It is the structural condition of operating at the frontier of AI in 2026: build responsibly, ship competitively, and convince public markets you can do both at the same time.
Enterprise teams that understand this dynamic are better positioned to make durable vendor decisions, rather than reacting to each quarterly model release as if it resets the board.
Sources: Reuters exclusive, September 19, 2026; InsideAI.news, September 19, 2026; SiliconReport, September 20, 2026; MarketScreener via Reuters, September 18, 2026