Prediction-market traders are betting that Anthropic will ship its next Claude Opus model within days, even as the company's own chief executive and a string of departing researchers press the artificial intelligence industry to ease off the accelerator.

Traders price in a late-September Opus

On Polymarket's "Next Claude Opus released by…?" market, the contract tied to a public release by September 24 was trading at around 83 cents on the dollar as of Monday, September 21, implying roughly an 83% chance of an on-time launch. Each cent traded on the platform roughly corresponds to a percentage-point probability, turning trader positioning into a live, crowd-sourced read on an outcome Anthropic itself has yet to confirm.

Confidence rises the farther out the deadline sits, a pattern that has held since the market opened on September 15. The market currently implies 94% odds of an Opus release by September 30, about 98% by October 31, and above 99% by year-end. It has drawn roughly $240,925 in trading volume so far. With the nearest deadline falling just three days after that Monday snapshot, the contract is set to deliver a fast verdict on the crowd's confidence.

Under the market's resolution criteria, a qualifying model must be a publicly accessible release explicitly named Opus, with an open beta or a rolling waitlist counting toward a yes. Other Anthropic model families, including Sonnet, Haiku, Fable, and Mythos, do not qualify.

Anthropic has not issued any official announcement confirming a release date or a version number for the next Opus model. Polymarket's own market page ties the late-September consensus to reports of stealth testing on a model codenamed claude-wafer-eap, which could be Claude Opus 5.5, though this remains unverified speculation.

Four Claude 5 models in under two months

Anthropic has been unusually busy with releases this year, which may explain why traders have begun opening markets on its future launches. The company shipped Claude Opus 5 on July 24 — its fourth Claude 5 model in less than two months — and then followed up with Fable 5.1 on September 1.

The acceleration is an industry-wide phenomenon rather than an Anthropic outlier. A Nikkei survey of five leading US labs and four Chinese firms, including Alibaba and Z.ai, found that the average gap between high-performance model updates has collapsed to 44 days for the April-to-September 2026 window, down from 125 days over the roughly three years through March 2026.

AI is writing most of Anthropic's code

Anthropic attributes the speed-up to AI increasingly building AI, and the company has published internal data to back that view. Data released on September 17 shows that Claude led about 26% of Anthropic's research and development work as of August, up from close to zero in February, with more than 90% of that work involving AI in some form.

According to company figures, Claude also authored more than 80% of the code merged into Anthropic's codebase as of May 2026, and a typical engineer merged roughly eight times as much code per day in the second quarter of 2026 as in 2024.

The Anthropic Institute sees the trend as a step toward recursive self-improvement — a loop in which AI systems increasingly build and refine AI — while cautioning that the milestone has not been reached and is “not inevitable.” OpenAI reported a parallel development, with AI agents logging 3.1 times the working hours of its human researchers in August.

The disclosures give hard numbers to a safety debate that has so far been waged largely through essays, resignations, and public warnings.

The people asking Anthropic to slow down

The pace of releases has ruffled feathers among those concerned about the capabilities of these models and how the labs are training them, with Anthropic at the forefront of most of the criticism.

CEO Dario Amodei published an essay titled “We Must Pace the Frontier on September 12, a call that drew public backing from OpenAI's Sam Altman and Elon Musk. His post came days after former Anthropic employee Jacob Coxon resigned on September 9 and warned that there is “a strong chance that we could all die in the immediate future.”

Bilal Chughtai, who left Google DeepMind's safety team in July, issued a similar warning. Microsoft AI CEO Mustafa Suleyman added his own critique on September 16, publicly calling out how Anthropic is training its Claude models. Suleyman described the approach as a dangerous mistake and said he is pushing for industry norms that would keep humans in control.

President Trump, for his part, does not appear to share the industry's concerns. He has dismissed the doom talk, saying his priority is beating China in the AI race. Whichever way the calendar breaks, the next Opus launch will land in the middle of an unusually public argument over how fast the frontier should move — and, if traders have priced it correctly, within days.

Source: Cryptopolitan