Anthropic's New Model Calculus Ahead of Its Planned IPO

Anthropic is weighing whether to introduce a new AI model in the run-up to its expected initial public offering, according to three people familiar with the matter, as the company seeks to blunt momentum gained by OpenAI after the launch of GPT-6 Astra this month. Anthropic is currently evaluating the safety of its next model as part of those release discussions, one person involved said.

The internal talks follow a very different public signal from Anthropic CEO Dario Amodei. On 12 September, Amodei published a 3,800-word essay calling on the AI industry to slow the pace at which model capabilities are improving, citing safety risks including autonomous AI agents escaping human control. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk both expressed support for the essay.

The competitive pressure is concrete. OpenAI released GPT-6 Astra on 3 September, and the model has won positive reviews from business users and developers. It accounted for about 13% of business AI spending tracked by expense platform Ramp, compared with about 8% for Anthropic's Claude Fable model. On OpenRouter, a widely used routing platform for developers, OpenAI models overtook Anthropic's models in user spending last week for the first time in more than two and a half years.

Anthropic still holds a revenue lead: its annualized revenue exceeded $65 billion by the end of July, up from roughly $9 billion at the end of 2025, and the company reportedly projects revenue of about $190 billion to $200 billion for 2028. OpenAI's annualized revenue passed $40 billion in July. Anthropic declined to comment on this story.

What GPT-6 Astra's Enterprise Gains Mean for Anthropic

Why Anthropic Would Launch Now Despite Its Safety Pause

The reported launch discussion puts Amodei's safety-first identity under direct strain. A release timed ahead of the IPO would be a commercial move aimed at OpenAI, not a cautious deceleration. That tension matters because investors and enterprise customers have treated safety as Anthropic's differentiator; if the company is seen as prioritizing listing momentum over its own safety position, it could weaken that brand advantage.

OpenAI's Real Enterprise Inroads — and Anthropic's Revenue Cushion

Ramp and OpenRouter data suggest OpenAI is closing the adoption gap among developers and smaller business purchasers. However, replacing entrenched enterprise vendors is slower, and Anthropic's $65 billion annualized revenue offers a much larger existing base than OpenAI's $40 billion. The key question is whether the newer spending data is an early indicator of churn or simply a surge from Astra's launch that will fade.

The Open-Source Threat That Outlasts the OpenAI Rivalry

Investors cited in the report identify a possibly bigger challenge: open-source models, particularly from China, can lower per-token costs and allow enterprises to run their own infrastructure outside Anthropic and OpenAI. Meta Platforms, once one of Anthropic's largest customers, is reportedly exploring ways to reduce its use of Anthropic models as it builds internal AI capability. That shift would pressure the economics of the commercial AI sector beyond the two-way rivalry.

What the IPO Timing Signal Means

OpenAI has removed some urgency by confirming it will not go public in 2026. Anthropic may move its own listing until after the US midterm elections in November, with marketing expected to start no earlier than mid-October. The delay gives Anthropic room to demonstrate that any new model can defend its enterprise position before investors price the offering.

Implications for Investors, Enterprise Buyers and AI Rivals

For investors, enterprise buyers and rival AI providers, the data points in this report matter most.

  • For investors: Track Ramp's business AI spending share and OpenRouter developer spending at least quarterly alongside Anthropic's next revenue disclosures. The expected IPO marketing window begins mid-October; a delay beyond the November US midterms would signal that Anthropic wanted more time to prove its next model against GPT-6 Astra before the listing.
  • For enterprise AI buyers: If you are contracting with Anthropic, ask for a written roadmap for the next model release before committing to a multi-year agreement. The shift from 8% to 13% Ramp share for OpenAI demonstrates that enterprise switching costs in this category are lower than many assumed, so contract exit clauses matter.
  • For AI providers: Meta's reported exploration of cutting Anthropic usage shows that even a marquee customer can move toward in-house AI. Calculate what share of your bookings is tied to customers with similar internal model programs; if meaningful, shorten renewal terms or add usage floors that survive partial self-supply.

Risk & Opportunity Assessment

Commercial RiskHighAstra's 13% Ramp share versus Claude Fable's 8% and the OpenRouter spending overtake create immediate commercial pressure, while open-source models threaten the token pricing behind Anthropic's 2028 forecast of $190-200 billion.
Competitive RiskHighOpenAI's GPT-6 Astra is gaining enterprise and developer adoption in third-party data; Anthropic's historic enterprise leadership is now being questioned by potential investors.
Regulatory RiskLowNo named regulatory or legal decision is reported; the only policy element is Amodei's voluntary safety position, not a binding rule.
Reputation RiskMediumLaunching a model shortly after the CEO publicly called for slower AI development risks a perception that Anthropic is prioritizing pre-IPO competition over its safety-first identity.
Technology DisruptionHighOpen-source models, including those from China, and Meta's internal AI efforts could reduce reliance on proprietary models and erode per-token pricing across the sector.
Commercial OpportunityHighA successful new release could blunt OpenAI's momentum and reinforce Anthropic's $65 billion annualized revenue base heading into an IPO after the midterm elections.