Cursor makes DeepSeek V4.1 Flash a selectable default at $0.60–$0.98 per task

Prasenjit Sarkar @stretchcloudOriginally posted 2026.10.27

Open original on X

A hundred million dollars a year, at a discount above thirty percent, is what it reportedly cost OpenAI to become a coding agent's default model instead of Anthropic. That's the Replit deal The Information reported today: most of Replit's AI spend moved from Anthropic to OpenAI, while users keep the option to pick Claude inside Replit Agent. Only the default changed. That's exactly the point, because almost nobody touches a dropdown once it's set. Within the same day, Cursor turned on DeepSeek V4.1 Flash as a selectable production default too, not a budget fallback buried in settings. Every effort level in their CursorBench run sits on the cost frontier, sixty cents to ninety eight cents a task, and nothing scores higher at that price. DeepSeek's V4.1 Flash runs as low as $0.0477 per million input tokens through API aggregators right now, roughly an 84 percent discount off the $0.30 Baidu Qianfan charges for the same capacity. Two platforms, same week, same conclusion. The model underneath a coding agent is now a configuration the platform negotiates on your behalf, priced like cloud compute, swapped when the economics move. Windsurf, Lovable and Bolt are running the identical playbook more quietly. Nobody markets "powered by GPT" anymore the way SaaS used to badge "powered by AWS." The model used to pick the agent. Increasingly the agent picks the model, on cost per task, not brand. If you're building on one provider with no abstraction layer between your product and the API call, you're the only party in this trade not getting a volume discount. https://x.com/cursor_ai/status/2114839997568405608

This post quotes another post: “DeepSeek V4.1 Flash in Cursor: $0.60–$0.98 per task on the CursorBench cost frontier”