AI Briefing: Kimi-K3 Tops Coding, DeepSeek Revenue & Model Safety Tests

Exponential View (Azeem Azhar) · rss · 2026-07-20

Key data signals across AI and markets: - **Kimi can code**: The newly released Kimi-K3 overtakes models like GPT-5.6 to take first place on the frontend code arena benchmark. - **Demand unlocked**: DeepSeek nears $500 million in annualized revenue, boasting gross margins as high as 70-80% on its V4 model. - **Ask and receive**: Safety tests on leading AI models using real terrorist prompts show a third of responses provided useful help to attackers. Labeling prompts as “research” increased model compliance from 17% to 42%.

Original post →

More from Venture

Venture channel →