Windsurf, the acclaimed vibe coding startup that is reportedly nearing acquisition by OpenAI, announced this week that its direct access to Anthropic’s widely used Claude AI models has been drastically reduced. In particular, the startup said it has lost substantial direct provisioned access to Claude 3.7 Sonnet and Claude 3.5 Sonnet, two AI models integral to Windsurf’s offerings.
According to CEO Varun Mohan, Anthropic provided Windsurf less than five days of notice before significantly cutting back on first-party compute capacity. Mohan expressed disappointment about the lack of advance communication, emphasizing Windsurf’s willingness and preference to pay for the full usage capacity directly from Anthropic.
This restriction forced Windsurf into rapidly sourcing third-party inference services in order to continue offering Claude-based functionality to users—a move the company acknowledged could lead to short-term model availability issues. While the company maintains some third-party compute resources, capacities are currently insufficient, and users may experience interruptions or degraded service levels until the alternative infrastructure scales up.
Anthropic’s decision appears part of a broader shift in priorities. The company recently launched Claude 4, a new generational AI model line highly targeted toward software engineering workflows, which was immediately available on several competing vibe coding platforms such as Cursor from Anysphere, Devin from Cognition, and GitHub Copilot from Microsoft. Windsurf was notably excluded from that initial rollout and has yet to receive direct access to the next-generation Claude models.
This move underscores additional friction between Anthropic and Windsurf. Earlier this year, Anthropic introduced its own AI-powered coding tool known as Claude Code, and subsequently hosted its inaugural “Code with Claude” developer conference aimed at positioning itself firmly within the competitive AI-driven developer tools market.
In response, Anthropic spokesperson Steve Mnich explained that the company is realigning its resources toward partnerships perceived as more sustainable and broadly beneficial for developer communities. He also pointed out developers still have alternatives, such as direct API access or tapping into Anthropic’s partner ecosystem.
Industry observers indicate Anthropic’s tightening of control over its AI models could hinder Windsurf’s otherwise rapid momentum. Windsurf’s revenue growth reached $100 million annual recurring revenue as recently as April, largely thanks to the industry’s rising adoption of AI-enhanced coding capabilities. As these powerful AI models have become essential to developers’ workflows, any access restrictions create immediate impacts. Users interviewed specifically noted difficulties when trying to leverage Anthropic’s latest AI offerings through Windsurf.
One user, Ronald Mannak, whose startup specializes in applications for Apple’s Swift language, recently moved his workflows away from Windsurf toward Cursor, citing easier availability and implementation of Anthropic’s advanced Claude 4 capabilities. To temporarily bridge the service gap, Windsurf gave users the option to use their personal Anthropic API keys within Windsurf, though developers criticized this approach as both costlier and more cumbersome compared to direct integration.
As OpenAI and other major players—including Google and Anthropic—consistently release improved AI models, companies offering AI-assisted development tools face critical challenges in guaranteeing continuous support and optionality. In an email statement, Windsurf reiterated its longstanding aim of providing its users substantial options but indirectly acknowledged that recent actions by Anthropic have complicated that mission considerably.
