Anysphere, the company behind the Cursor AI coding environment, issued an apology Friday regarding a poorly communicated pricing adjustment to its $20-per-month Pro plan, which led to unexpected costs for some users.

Michael Truell, CEO of Anysphere, stated in a blog post, “We recognize that we didn’t handle this pricing rollout well and we’re sorry. Our communication was not clear enough and came as a surprise to many of you.” This statement refers to a June 16 update for Cursor’s Pro plan. Previously, the plan offered 500 fast responses using advanced AI models from OpenAI, Anthropic, and Google, followed by unlimited responses at a slower speed. Under the revised plan, subscribers now receive $20 worth of usage per month, billed at API rates. This new structure permits users to execute coding tasks within Cursor using their preferred AI model until the $20 limit is reached, after which additional credits must be purchased for continued use.

Following the announcement, Pro users voiced complaints on social media platforms. Many reported exhausting their allocated requests rapidly under the new plan, sometimes after initiating only a few prompts, particularly when utilizing Anthropic’s Claude models, which are frequently employed for coding tasks. Other users claimed they incurred unanticipated additional charges due to not fully comprehending that exceeding the $20 usage limit would result in extra fees if a spending cap had not been established. The updated plan only offers unlimited usage for Pro users through Cursor’s “auto mode,” which automatically routes requests to AI models based on available capacity.

Why is no one talking about how @cursor_ai essentially rug pulled???

Making the pricing “unlimited” and then changing it to limited without any warning at all, i check dashboard and they bill me $71 for just 1 day of sonnet-4 usage???

— 🦇عثمان (@0ni_x4) July 3, 2025

Anysphere has stated its intention to refund users who were unexpectedly charged and aims to enhance clarity regarding future pricing modifications. Truell noted in the blog post that Anysphere revised Cursor’s pricing because “new models can spend more tokens per request on longer-horizon tasks.” This indicates that certain contemporary AI models have become more expensive due to their consumption of significant computational resources and time to complete intricate, multi-step tasks. Cursor had absorbed these expenses under its previous Pro plan but is now transferring these costs to its users.

While the cost of many AI models has decreased, the expense associated with cutting-edge performance remains substantial, in some instances exceeding previous levels. Anthropic’s recently introduced Claude Opus 4 model is priced at $15 per million input tokens, which equates to approximately 750,000 words, a length surpassing the entire “Lord of The Rings” series. Output tokens for the model are priced at $75 per million.

This pricing structure is even higher than Google’s Gemini 2.5 Pro, which was launched in April and represented its most expensive AI model at the time. In recent months, OpenAI and Anthropic have also initiated charges for “priority” access to AI models for enterprise customers. This represents an additional premium beyond standard AI model costs, ensuring reliable and high-speed performance.

Cursor launches web app for coding agents

These increasing expenses appear to be influencing the pricing of AI coding tools across the industry. Users of Replit, another widely used AI tool, also encountered unexpected pricing changes recently, which increased the cost of completing large tasks with AI assistance. Cursor has achieved significant market success, reaching over $500 million in Annual Recurring Revenue (ARR) primarily through subscriptions to its Pro plan. However, Cursor currently faces considerable competition from the very AI providers it relies upon, while simultaneously seeking methods to affordably integrate their more expensive AI models.

Anthropic’s recently released AI coding tool, Claude Code, has gained traction among enterprises, contributing to a reported increase in Anthropic’s ARR to $4 billion and potentially drawing users away from Cursor. Last week, Cursor recruited two Anthropic employees who were instrumental in the product development of Claude Code.

For Cursor to maintain its leading market position, continued collaboration with state-of-the-art model providers is necessary, at least until its own internally developed models achieve competitive viability. Consequently, Anysphere recently finalized multi-year agreements with OpenAI, Anthropic, Google, and xAI to introduce a $200-a-month Cursor Ultra plan, which offers very high rate limits. Jared Kaplan, co-founder of Anthropic, confirmed to TechCrunch in June his intention to maintain a long-term working relationship with Cursor.

Featured image credit