In brief: Anthropic is considering a new AI model to counter OpenAI's GPT-6 Astra momentum, Reuters reports. A release, date and technical details are not confirmed. The tension is between Anthropic's safety pacing and competitive and IPO pressure.
What is reported
Three sources told Reuters that Anthropic is weighing a rollout; one said safety evaluations are under way. Anthropic declined to comment. It is neither decided that a model will ship nor established what capabilities, name or access model it would have.
Reuters puts Anthropic's annualised revenue run rate above $65 billion in July, versus more than $40 billion for OpenAI. Ramp data it cites tracked about 13% of spending for Astra and 8% for Claude Fable. A separate OpenRouter analysis reportedly put OpenAI ahead of Anthropic last week for the first time in more than two and a half years. A possible IPO could slip until after the US elections in November.
Pandorex Analysis
The deliberations do not yet contradict Anthropic's safety rules. Its Responsible Scaling Policy does not promise a release freeze; it specifies risk-based evaluations, safeguards and public Risk Reports above defined thresholds. The key test is whether Anthropic documents model risks, access and safeguards before release. Until then, assessing the unknown model is speculative.
The market signals are not equivalent to market share. Ramp measures spending among a particular customer set. OpenRouter's public ranking measures processed tokens, lists model variants separately and does not automatically reproduce the spending comparison reported by Reuters. Quality, token use, revenue and enterprise adoption can consequently produce different rankings.
The defensible connection is timing. Days ago, Anthropic CEO Dario Amodei warned about overly rapid progress and called for embedded external evaluators. The company is now reportedly weighing a faster response alongside its IPO plans. That does not prove a reversal, but raises the credibility cost of any shortened or opaque review.