Media Zone | 2026-07-09

Grok 4.5 dominated the timeline. xAI engineers celebrating, Artificial Analysis confirming frontier tier, and the coding-agent angle everywhere.

Today's signal

• Dominant story: Grok 4.5 hit the intelligence frontier (index 54, 4th overall), coding-parity with GPT-5.5.
• Pattern: xAI engineers all crediting RL scaling and "intelligence per token" as the lever.
• Counter-signal: numbers are from xAI and Artificial Analysis, no independent harness-normalized confirmation.
• Quiet area: no research-paper social chatter today, all product and benchmark noise.

LLMs, agents, safety

Grok 4.5 reaches the frontier

• SpaceXAI's first coding-and-agents model scored 54 on Artificial Analysis Index, 4th overall.
• Coding-parity with GPT-5.5 in the Grok Build harness, at much lower cost.
• xAI engineers credit RL scaling, RL stability, and intelligence-per-token gains.
• 16-point jump over Grok 4.3, outperforms all open-weight models.

Industry and business

• Three humanoid robotics companies moved toward IPO this week (AI Weekly robotics edition).
• Google at ICML 2026 Seoul: Gemini Robotics, 3D Transformers, Isomorphic Labs drug design.
• Anthropic extended Fable 5 subscription access to July 12 as capacity secured.