AI Industry Highlights: OpenAI Shake‑up, China’s Model Surge, Gemma 4 Open‑Source, and Cursor 3

01 Today’s AI Highlights

OpenAI leadership turmoil : The CMO Kate Rouchi, AGI deployment lead Figi Simo (on medical leave), and COO Brad Lightcap all left their posts on the same day, a move far beyond market expectations. Lightcap, a trusted operator behind ChatGPT’s global rollout and API commercialization, departing just before an IPO suggests possible restructuring, but also raises concerns about management stability.

OpenAI’s economic reform proposal : The company released a policy paper outlining four‑day work weeks, a revised tax system shifting focus to corporate profit and capital gains, a universal basic income funded jointly by governments and AI firms, and accelerated grid infrastructure to support future compute and energy needs. OpenAI admits the proposals are exploratory and aim to shape regulation in the emerging super‑intelligence era.

Google Gemma 4 open‑source : Released on April 2, Gemma 4 is a small‑parameter model that scores 85% of the performance of a competitor model 20× larger on BigBench reasoning and coding benchmarks. It supports a 256K context window, includes a dedicated reasoning mode for complex multi‑step logic, and, thanks to NVIDIA collaboration, reduces local inference compute requirements by roughly 40%.

Cursor 3 launch : The new generation moves from a Copilot‑style code completion tool to an “agent” that can autonomously decompose tasks, write, test, and debug code. Features include an “agent mode” where users describe requirements and the AI handles the entire workflow, plus real‑time collaboration and self‑healing debugging.

OpenAI, Anthropic & Google partnership : Reported by Bloomberg, the three firms are strengthening cooperation through the Frontier Model Forum, sharing safety signals to detect and prevent “adversarial distillation” that could copy advanced model capabilities.

02 Global AI Dynamics

OpenAI leadership shake‑up : Interpreted as a signal ahead of the IPO, reflecting a shift from a pure technology company toward a broader societal infrastructure role.

03 China AI Dynamics

OpenRouter data shows global weekly AI model usage reached 27 trillion tokens, up 18.9% week‑over‑week.

China’s weekly usage hit 12.96 trillion tokens, a 31.48% increase from the previous week, surpassing the United States, which logged 3.03 trillion tokens (0.76% growth).

The top six models by global usage are all Chinese products.

Alibaba Qwen 3.6 series :

Qwen 3.6 Plus (free) achieved 4.6 trillion tokens, topping the global leaderboard.

Qwen 3.6 Plus Preview recorded 1.64 trillion tokens, ranking third.

Other notable models: Xiaomi MiMo‑V2‑Pro (3.08 trillion), Step 3.5 Flash, MiniMax M2.7, DeepSeek V3.2.

In the Code Arena React benchmark, Alibaba’s Qwen 3.6‑Plus placed second, overtaking OpenAI and Google, highlighting Chinese models’ growing competitiveness in programming tasks.

04 Today’s Summary

Strategic divergence among top firms : OpenAI’s leadership changes and policy proposals signal a move toward positioning AI as societal infrastructure, while Google pursues an open‑source ecosystem with Gemma 4.

China’s scale advantage : Token‑level usage data demonstrates Chinese models’ cost‑performance and usability advantages translating into tangible market share.

AI programming enters the agent era : Cursor 3’s release marks a shift from simple code completion to autonomous AI‑assisted development, urging developers to adapt to collaborative workflows.

Model distillation becomes a focal challenge : Joint efforts by leading AI companies aim to protect core model capabilities, potentially reshaping open‑source strategies and the roadmap for smaller players.