LWiAI Podcast #252 - GPT 5.6, Grok 4.5, Nemotron-Labs-Diffusion, AI 2040

Last Week in AI released its 252nd podcast episode, recorded on 07/11/2026 and hosted by Andrey Kurenkov and Jeremie Harris. The episode summarized and discussed the week's AI news.

The episode covered OpenAI publicly rolling out GPT-5.6, including Sol and Luna, and rebranding its desktop agentic coding product as ChatGPT Work. It noted disputed claims about whether the US government effectively green-lit and delayed the release, along with concerns about inconsistent, ad hoc frontier-model oversight and jailbreakability.

On model releases, SpaceX AI's Grok 4.5 launched as a very low-cost, Opus-class coding model with minimal safety documentation, per the episode. Meta released Muse Spark 1.1 with aggressive pricing, large coding and cyber benchmark gains, and a lengthy safety evaluation. Meta also previewed Muse Video and rolled out Muse Image before backtracking after backlash over easy generation of images of public Instagram accounts. Chinese open-source models grew to over 30% of weekly OpenRouter tokens as cost pressure increased, alongside discussion of risks such as potential insider threats.

Infrastructure, policy, and safety items included Meta exploring selling AI compute as a cloud business, US energy regulators pressing grid operators on large-load data-center connections, and Anthropic publishing a "global workspace" interpretability method for verbalizable internal representations. Reports indicated China may restrict overseas access to top models, and AI 2040 proposed US-China coordination to slow progress until alignment improves.

Open-source and project segments included Nemotron-Labs-Diffusion, described as a tri-mode language model unifying autoregressive, diffusion, and self-speculation decoding, and Tencent's Hy3, an open 295B mixture-of-experts model with 21B active parameters and 256K context. In a note, host Andrey Kurenkov said he was resuming posting, catching up on missing episodes, and had paused paid subscriptions until posting is consistent.