Which AI Coding Tool Wins? A Hands‑On Benchmark of Cursor, DeepSeek, and Doubao

Introduction

With the rapid rise of artificial‑intelligence technologies, AI programming tools have emerged that can understand natural‑language requirements, generate and optimise code, and assist debugging, dramatically improving development efficiency. This test evaluates Cursor + Claude 3.7, DeepSeek, and Doubao AI programming under a unified prompt to provide developers a clear reference for tool selection and to encourage tool developers to optimise their products.

Test Tools

Cursor + Claude 3.7

Cursor is a revolutionary AI programming tool that has been deepening its AI coding capabilities since November 2023. Initially relying on GPT‑4 and DeepSeek, it added Claude 3.7 support in March 2025 and achieved high performance with its Max mode, positioning itself as an “AI programming operating system” especially suitable for complex enterprise projects.

DeepSeek

DeepSeek has been continuously advancing in AI programming since November 2023, progressing through Coder, V2, V3 to V3‑0324. Its March 2025 release, DeepSeek‑V3‑0324, represents its current top‑tier coding ability, is free and open‑source, and can run on ordinary computers, offering a high‑value option for developers, students, and startups.

Doubao AI Programming

Doubao launched its AI programming feature on desktop and web in January 2025, supporting batch file uploads, real‑time GitHub repository integration, and a new code editor with immersive reading. In June 2025 it added a visual‑programming “Application Creation 1.0” feature allowing drag‑and‑drop web‑app creation.

Test Plan

Environment

Hardware: iMac Intel Core i5 processor, 20 GB RAM, 1 GB SSD.

Software: macOS Ventura 13.64, Cursor 1.1.6 (trial), DeepSeek web version, Doubao Chromium Engine 129.0.6668.79 (official).

Prompt

The prompt outline from Bilibili creator “AI Evolution Theory‑Peanut” was adapted by changing the keywords to “financial product” (for learning and exchange only, no commercial use).

Metrics

Generation efficiency: total time from prompt input to complete page rendering.

Functional completeness: coverage of core business processes and completeness of implemented features.

Visual presentation: assessment of modern UI design, visual hierarchy, and aesthetic quality of interactive elements.

Test Process

Cursor + Claude 3.7

After entering the prompt, Cursor performed three steps—analysis, file creation, and summarisation—to generate pages. Six pages (home, loan, loan detail, personal centre, authentication centre, repayment) were produced in 11 min 23 s, using a tiled layout with vertical scrolling for content beyond a single screen.

DeepSeek

DeepSeek followed three steps—deep thinking, code implementation, and design—to generate a high‑fidelity prototype of five core pages (home, loan, billing, my, submission) in 7 min 12 s. Navigation was handled via a bottom tab bar, and the prototype supported clicks, swipes, and selections to simulate real‑world usage.

Doubao AI Programming

Doubao generated four pages (home, loan, billing, my) in 2 min 51 s. The prototype could be downloaded and opened locally; navigation tabs extended beyond the mobile emulator frame, but the core functionality met the test criteria.

Conclusion

Based on the benchmark, Doubao AI Programming achieves the fastest generation speed, making it the preferred tool for rapid single‑page prototyping, especially for beginners. DeepSeek excels in visual design and interaction richness, suitable for scenarios demanding refined user experience. Cursor, while slower and with a higher entry barrier, offers flexible custom editing capabilities that cater to experienced developers needing deeper control.