Invest1 publisher3 min readPublished Updated
Traders pay 83 cents for an Anthropic Opus release nine days after its CEO urged pacing
Anthropic says Claude led about 26% of its R&D in August and wrote more than 80% of the code merged into its codebase, and a thin prediction market gives its next Opus an 83% chance of shipping by September 24.
The Investor · Invest desk
What happened
- Polymarket's contract on a publicly released Anthropic model explicitly named Opus arriving by September 24 traded around 83 cents on the dollar on Monday, September 21.
- The same disclosure put Claude's share of code merged into Anthropic's codebase above 80% as of May 2026.
- A Nikkei survey of five US labs and four Chinese firms found the average gap between high-performance model updates fell to 44 days for April to September 2026 from 125 days.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- constraint At a 44-day average gap, a quarter holds about two high-performance updates across the surveyed labs, so a single voluntary pause is priced in competitor releases rather than in months.
- contradiction Anthropic attributes its faster release cycle to AI building AI while its chief executive asks the industry to pace the frontier, and so far as the record shows it has kept to its own schedule.
- exposure At about $40,000 of daily volume, the 83-cent price is set by a handful of traders, so anyone treating it as a proxy for Anthropic's internal calendar is reading a very small book.
- capability Eight times the merged code per engineer per day means a lab can hold headcount flat and still raise output, which weakens hiring as a read on how fast a lab is moving.
The internal structure of the Polymarket book is more informative than its headline price. The September 24 contract sits near 83 cents and the September 30 contract at 94, so traders are assigning 11 points to the six days in between [1]. The move from 94 to about 98 by October 31 leaves the whole of October carrying four points [2].
It is also a small book. Roughly $240,925 has traded since the market opened on September 15 [3], which across the six days to Monday works out near $40,000 a day [3]. A buyer at 83 cents risks 83 to collect a dollar, a gross return of about 20% if the release lands inside the window [4]. Anthropic has not confirmed a date or a version number, and Polymarket's page attributes the late-September consensus to reported stealth testing of a model codenamed claude-wafer-eap, which cryptopolitan describes as unverified speculation [5][6].
Five days after Dario Amodei published "We Must Pace the Frontier" [14], Anthropic put out its own numbers. Claude led about 26% of its research and development work as of August, up from close to zero in February [9][9]. Six months, 26 points, so roughly 4.3 points a month [5]. Anthropic's own explanation for the faster release cycle is that AI is increasingly building AI [19].
Only one reading of that pair is a contradiction. Pacing the frontier may refer to the size of each capability jump. On that reading, shipping an Opus increment weeks after Claude Opus 5 on July 24 is consistent with the essay [7], and the article gives only the title of what Amodei wrote. The 26% and the 80% are throughput figures. A typical engineer merging roughly eight times as much code per day in the second quarter of 2026 as in 2024 [11] tells you about volume inside one codebase. The 44-day average update gap covers five US labs and four Chinese firms including Alibaba and Z.ai [8], so it can collapse because Beijing ships fast.
Even so, 44 against 125 is the figure I would weight [8]. That average covers April to September 2026, the window containing both Amodei's essay and the public backing it drew from Sam Altman and Elon Musk [14]. It came in at less than a third of the prior three-year pace [6].
The safety case is being made by people on the way out and by rivals. Jacob Coxon, who left Anthropic on September 9, said there is "a strong chance that we could all die in the immediate future" [15]. Bilal Chughtai left Google DeepMind's safety team in July [16]. Mustafa Suleyman, Microsoft AI's chief executive, said on September 16 that Anthropic's method of training Claude is a dangerous mistake and pushed for industry norms that keep humans in control [17]. The Anthropic Institute, describing the automation figures as a step toward recursive self-improvement, said the milestone is "not inevitable" [12].
I would change my view if Anthropic attached dates to a pacing commitment, or if the next Nikkei reading showed the average gap widening from 44 days. The September 24 contract resolving no would leave that view intact, because the same book already prices a release by year-end above 99% [2].
What to watch
- Whether the September 24 contract resolves yes, and what the 94% September 30 line does if it does not.
- The next Nikkei reading on the average gap between high-performance model updates, following the 44-day figure for April to September 2026.
- Whether Anthropic attaches dates or thresholds to what Amodei means by pacing the frontier.