Version 1

SemiAnalysis ran empirical tests on top-tier Anthropic and OpenAI subscription plans — purchasing each and running long-horizon coding tasks until hitting weekly limits — and found that for heavy users, subscriptions are 40–70x cheaper than paying equivalent API rates [3][2]. The Wall Street Journal separately reported that OpenAI is weighing deep price reductions in response to Anthropic's competitive pressure, concentrated in developer and coding workflows through Claude Code [6][7]. Anthropic's valuation has reached approximately $965 billion, surpassing OpenAI's [11].

The 40–70x subscription-to-API cost gap shows that current pricing is strategically generous to heavy individual users, likely to drive adoption rather than maximize near-term margin. As agentic coding tools generate unusually high token volumes per session, subscription economics become a live competitive variable — and OpenAI's reported consideration of price cuts suggests the pressure is forcing responses rather than being absorbed quietly.

SemiAnalysis published a four-part analytical thread on June 10, 2026, comparing subscription and API business models for AI labs [1]. The core empirical contribution: the firm purchased top-tier subscriptions from both Anthropic and OpenAI and ran long-horizon coding tasks until hitting weekly usage caps, then compared the implied token cost to API rates for the same volume [2]. The finding, amplified by Rohan Paul, is that heavy coding and chat users on subscriptions pay 40–70x less than they would via the API; the API's primary advantage is for automation and product integration use cases [3].

SemiAnalysis frames subscription margin as a function of average utilization across the subscriber base [4]. Assuming 75% API gross margins for both Anthropic and OpenAI, the actual subscription margin depends on how aggressively users consume their allowance. The analysis notes that explicitly throttling subscriptions would generate significant public backlash, and argues that declining inference costs will eventually close the economics gap — making it possible to profitably serve frontier-class models at $20/month without throttling [5]. The thread is structured as a deliberate teaser, withholding the specific utilization figures while establishing the analytical framework.

On June 11, the Wall Street Journal reported that OpenAI is considering deep price reductions, citing competitive pressure from Anthropic as the driver [6][7]. Anthropic's growth is concentrated in developer and coding workflows, particularly Claude Code, where users generate unusually large token volumes per session — making pricing especially consequential for both companies' revenues [6]. Multiple outlets picked up the WSJ report [8][9][10].

A separate WSJ report noted that Anthropic's valuation has reached approximately $965 billion, surpassing OpenAI [11]. Together, these data points describe a market where Anthropic has gained ground in high-value developer segments, OpenAI is under pressure to respond on price, and both companies face the structural question of how to convert heavy subscription users into sustainable margin — a problem SemiAnalysis argues falling compute costs will eventually resolve on their own.