Astra seized 13% of the market share in two weeks, and Anthropic may release its new model ahead of the original schedule.

Edited by Shan Hui 

Anthropic keeps talking about "slowing down", but is about to release a new model soon after?

Reuters just published an exclusive report: Anthropic is considering releasing its new model ahead of schedule.

It seems that GPT-6 Astra has really dealt a heavy blow to it.

Two weeks, reshaping the market pattern built in two and a half years

On September 1, Anthropic just launched Fable 5.1, which was officially positioned as "the most powerful model for programming and knowledge work". It also voluntarily cut the cache reading price by 75%, trying to retain enterprise customers with cost-effectiveness.

It never expected that OpenAI would follow up to compete. On September 3, OpenAI released Astra. Sam Altman said in his own words: "We are working as hard as we can to get Astra into everyone's hands as soon as possible."

This time it is far more than just "feeling the pressure" — Anthropic is witnessing its market share erode in real time.

Statistics from OpenRouter show that Astra has been online for just over two weeks, and has captured about 13% of the enterprise AI spending share, while Anthropic's Fable series only accounts for 8%.

Peter Walker, Head of Analytics at OpenRouter, directly released several charts showing the whole process of Astra starting from zero and gradually squeezing down Anthropic's market share.

According to official OpenRouter data, the weekly token usage share of OpenAI and Anthropic's models from January to September 2026: the orange line represents Anthropic, the blue line represents OpenAI, and the 50% dashed line is the dividing line. From January to June, Anthropic steadily held a share above 50%; after Astra went online, OpenAI's share climbed rapidly, reaching about 65% in the first week of September, while Anthropic's share dropped to about 35%.

For the first time in two and a half years, OpenAI surpassed Anthropic.

Calling for "deceleration" while stepping on the accelerator

On September 12, Dario Amodei, CEO of Anthropic, published a long article publicly calling for "slowing down the development speed of AI".

Only seven days have passed since then.

Although both Altman and Musk publicly echoed his point of view rarely, saying "Dario is right", the "slowdown" that Anthropic can accept is probably a "slowdown" when it still holds a leading position.

Seeing that its competitors did not slow down, Anthropic could no longer stay calm. Although the safety narrative has long been the brand slogan of Anthropic, commercial competition does not follow that logic.

Two months after the release of Fable 5, enterprise customers did not actively upgrade to the most expensive flagship model as expected, and the share stayed at around 11% without further growth. Enterprise customers voted with their feet and chose the cheaper but "sufficiently capable" Opus 5.

Now with the arrival of Astra, even the "sufficiently capable" position may be taken away.

Releasing the new model in advance is a commercially reasonable defensive move.

The IPO plan is pushing Anthropic to move forward, and the company that once took "responsible AI" as its banner is unexpectedly forced to evaluate whether to disrupt its original release rhythm within two weeks to respond to the "acceleration race" that it has publicly criticized.

The Reuters report also pointed out that Anthropic's IPO may be postponed to November. Sources also said that Claude is deeply involved in the R&D of the next-generation model.

Obviously, in the countdown to the IPO, the growth curve is more important than the safety declaration.

Reference links:

https://www.reuters.com/business/anthropic-considers-releasing-new-ai-model-ahead-ipo-sources-say-2026-09-19/

© THE END 

This article is from the WeChat public account "jiqizhixin" (ID: almosthuman2014), authored by AI-focused contributors, and republished with authorization from 36Kr.