Why 1.4 B Tokens Only Puts Me in the Top 25% of Cursor Users While Team Leaders Consume 9.1 B?

Cursor released its 2025 personal usage report, and the author examined their own statistics out of curiosity. Despite using multiple AI coding tools—including Gemini CLI, Trae International and domestic versions, OpenAI Codex, and Claude Code—their Cursor usage still dominates, placing them in the Top 25% of active users.

Key personal metrics:

Joined 361 days ago

Usage rank: Top 25%

Agent dialogue count: 6 K

Opened tabs: 12.8 K

Tokens consumed: 1.41 B

The three most used models are Claude 4 Sonnet, Claude 3.7 Sonnet, and Claude 3.5 Sonnet. Claude has been the primary model because of its stable code‑understanding, context handling, and generation quality. Although Claude 4.5 was tried, it saw fewer interactions, while Codex‑based GPT‑5 models were also heavily used for complex problems.

Heavy‑User Data from the Cursor Team

Eric Zakariasson, a Cursor team member, shows a "god‑level" usage profile:

Usage rank: Top 0.1%

Joined 396 days ago

Agent dialogue count: 45 K (7.5 × the author)

Opened tabs: 11.6 K

Tokens consumed: 9.10 B (6.5 × the author)

Consecutive usage days: 151 (half a year daily)

Top three models for Eric are Claude 4.5 Sonnet, Claude 4 Sonnet, and GPT‑5.

Lee Robinson – Vercel VP of Product

Lee Robinson’s six‑month usage also stands out:

Joined 175 days ago

Agent dialogue count: 4.7 K

Opened tabs: 2.6 K

Tokens consumed: 2.03 B (20 × 10⁸)

Consecutive usage days: 42

His top three models are Claude 4.5 Opus (dominant token consumer), Composer 1, and Claude 4 Sonnet. He noted in a tweet that he will reward the user who spends the most tokens building projects with a large amount of Cursor credits.

Ryo Lu – Full‑Stack Model User

Ryo Lu’s annual trend chart shows near‑complete coverage of models supported by Cursor, including the full Claude series (3.5, 3.7, 4, 4.5), GPT‑5, Gemini 2.5 Pro, o3, and Composer 1. The chart illustrates that usage peaks shift over time, indicating that users switch models flexibly based on task requirements rather than relying on a single model.

Overall, the data suggest that heavy users treat AI coding assistants as "personal programming partners" for rapid idea construction and validation. The author concludes that while AI tools reshape workflows, they are amplifiers—not substitutes—for human judgment.