Matt Wolfe surveys a week of AI news, separating his hands-on model tests from announcements and speculation. His OpenAI Decisions API segment is sponsored.

Nvidia’s leadership change

  • Wolfe discusses the reported investigation, CEO transition and potential implications for Nvidia’s role in AI infrastructure.
  • His predictions about what happens next are speculation; the personal and chip-sales allegations he discusses should not be treated as established findings.

Model comparisons

  • Wolfe says Claude Fable 5.2 is his current daily driver; his visual benchmark placed it seventh and cost about $5 for that particular generation.
  • In his 3D game test, he preferred Gemini 4 Carbon’s visual result; his model-judged visual test placed Carbon second. Those are his tests, not general capability rankings.
  • Gemini 4 Hydrogen targets a smaller, faster and cheaper tier; leaderboard scores and task costs depend on the benchmark and configuration.

Tools and interfaces

  • In a sponsored segment, Wolfe describes OpenAI’s Decisions API for probability, choice and scoring tasks; its speed and pricing are vendor claims.
  • Adobe showed AI-assisted editing and experimental previews at MAX; Wolfe says many ideas resemble capabilities already seen elsewhere.
  • He also rounds up business profiles in ChatGPT, Meta’s Muse Ember, Google’s Decisions API, Claude’s sheets and collaborative tools, MiniMax image generation and Automattic’s Spacefast.
  • Several items are announcements or demos rather than products he independently tested; access, pricing and availability may change.

“I actually think this did a better job than Fable 5.2 did with the game.” — Wolfe on his Gemini 4 Carbon test