This Week in AI: GPT-5.6, Grok 4.5, and the EU AI Act Countdown

Frontier labs spent the week in a release sprint: OpenAI unveiled its GPT-5.6 family and a new voice stack, Meta and xAI pushed deeper into agentic coding, Google opened AlphaEvolve to all customers, and Tencent dropped a major open-weight model. Meanwhile, regulators in Brussels and Washington set the compliance tone for the months ahead.

Models

GPT-5.6: Frontier intelligence that scales with your ambition — OpenAI launched Sol, Terra, and Luna on July 9, positioning Sol as a new flagship with stronger coding, cybersecurity, and agentic benchmarks plus programmatic tool calling in the Responses API.

Introducing Grok 4.5 — xAI released Grok 4.5 on July 8 as its top coding and knowledge-work model, emphasizing fast inference, roughly 2× token efficiency, and availability in Grok Build, Cursor, and the API.

Introducing Muse Spark 1.1 — Meta upgraded its multimodal reasoning model for agentic tasks on July 9, with gains in tool use, computer use, and coding, and opened a public preview of the Meta Model API for developers.

Tencent Hunyuan Officially Releases Hy3 — Tencent open-sourced Hy3 on July 6, a 295B-parameter MoE agent model with Apache 2.0 licensing, 256K context, and integration across WeChat, Yuanbao, and CodeBuddy products.

Introducing Muse Image and Muse Video — Meta also debuted Muse Image on July 7 for instruction-following image generation and editing, with Content Seal watermarking, and previewed Muse Video with native audio support.

Agents & Tools

Introducing GPT-Live — OpenAI rolled out GPT-Live on July 8, a full-duplex voice model for ChatGPT that can keep talking while delegating harder tasks to frontier models like GPT-5.5 in the background.

AlphaEvolve is available for everyone on Google Cloud — Google made AlphaEvolve generally available on July 9 on the Gemini Enterprise Agent Platform, offering an optimization agent that has already improved TPU design, Spanner compaction, and compiler strategies internally.

Policy & Regulation

EU AI Act this week: the Council seals the Omnibus — The Council gave final approval to the Digital Omnibus on AI, keeping the 2 August 2026 enforcement date for general-purpose AI rules and Article 50 transparency duties while pushing some high-risk deadlines to 2027–2028.

FTC Policy Statement on Suppression of Accuracy in AI Systems — The U.S. Federal Trade Commission proposed on July 7 that steering AI outputs away from user expectations without disclosure may violate Section 5, with comments due by 31 July.