Anysphere CEO Addresses Controversial Price Changes for Cursor Pro Subscription The CEO of Anysphere, the developer behind the popular AI-powered coding tool Cursor, apologized to users on Friday following complaints about recent price changes to its $20/month Pro subscription. The adjustments led to unexpected charges for some customers, prompting backlash on social platforms. "We acknowledge missteps in our communication during this pricing transition and sincerely regret the confusion caused," wrote CEO Michael Truell in an official blog post. "Our messaging lacked clarity, leaving many users unprepared for these changes." The updates, implemented on June 16, restructured the Pro plan's token allocation system. Previously, subscribers received 500 fast responses across OpenAI, Anthropic, and Google's advanced AI models followed by slower-rate unlimited access. Under the new structure, users now receive $20 worth of AI model usage (billed at API rates) per month. This allows coding tasks in Cursor with selected models until the $20 threshold is reached, after which additional credits must be purchased to continue usage. Social media reactions quickly emerged in the weeks following the announcement. Many users reported exhausting their monthly allowance rapidly, with some encountering depleted credits after just a few prompts using Anthropic's Claude models—widely adopted for coding. Others expressed surprise at unexpected charges when exceeding the $20 limit without setting spending caps. Notably, only the "Auto Mode" feature now provides Pro users with unlimited access by automatically routing requests to available models based on capacity. Anysphere pledged to refund affected users and enhance future communication regarding pricing updates. The company declined to comment further beyond the blog statement provided to TechCrunch. Truell explained in the blog that the pricing shift stems from "newer models consuming significantly more tokens over extended tasks"—a cost increase for complex multi-step operations now passed to users. While Cursor previously absorbed these expenses under its old pricing model, the company now prioritizes cost transparency for consumers. This trend mirrors broader industry pressures as leading AI providers continue raising costs. Anthropic's recently launched Claude Opus 4 model charges $15 per million input tokens (equivalent to 750,000 words or the entire *Lord of the Rings* trilogy) and $75 per million output tokens, outpricing even Google's $19/million input/output Gemini 2.5 Pro model. Both companies have also introduced premium "priority access" fees for enterprise clients to guarantee reliable high-speed performance—a markup layered atop existing service costs. These increasing expenses are cascading through the AI tool ecosystem. Competitor Replit recently faced similar user backlash over sudden price adjustments, making large-scale AI tasks significantly more expensive. Cursor, having generated over $500 million in annual recurring revenue primarily through Pro subscriptions, now faces dual challenges: competition from its own model providers and the need to affordably deliver rising-cost AI capabilities. Anthropic's enterprise-focused coding tool Claude Code has attracted both business users and reportedly some Cursor subscribers, boosting its ARR to $4 billion while luring key personnel from Anysphere last week. To maintain its market leadership, Cursor must continue partnering with cutting-edge model developers until its own models achieve competitive parity. This strategy has materialized in Anysphere's recent multi-year agreements with OpenAI, Anthropic, Google, and xAI to offer the $200/month Cursor Ultra plan with high-rate limits. Anthropic co-founder Jared Kaplan also affirmed long-term collaboration with Cursor in a June interview with TechCrunch. However, tensions between coding tools and model developers remain evident as the industry navigates escalating infrastructure costs.