LWiAI Podcast #252: OpenAI Launches GPT-5.6, LLM Pricing War Intensifies

Last Week in AI · rss · 2026-07-21

Last Week in AI Episode 252 returns with a comprehensive breakdown of recent, fast-paced developments in the AI industry.

Model & Product Releases

OpenAI: Publicly rolled out GPT-5.6 (including Sol and Luna) and rebranded its desktop agentic coding product as ChatGPT Work. The release was marred by disputes over whether the US government effectively delayed it, raising concerns about ad hoc frontier-model oversight.
xAI & Meta: xAI launched Grok 4.5, a low-cost, Opus-class coding model with minimal safety documentation. Meta released Muse Spark 1.1 with aggressive pricing and large coding/cyber benchmark gains, and previewed Muse Video/Image, quickly backtracking after backlash over the easy generation of public figures.

Industry Trends & Open Source

Rise of Chinese Models: As cost pressures mount, Chinese open-source models now account for over 30% of weekly OpenRouter tokens, posing a significant threat to top closed-source labs.
Infrastructure: Meta is exploring a cloud business to sell AI compute. Concurrently, US energy regulators issued an ultimatum to grid operators over massive power demands from data centers.

Safety, Policy & Research

Policy: Sources suggest Beijing is considering restricting overseas access to China's top AI models. The ex-OpenAI employee behind 'AI 2027' proposed the 'AI 2040' plan, recommending US-China coordination to slow progress until alignment improves.
Interpretability & Architectures: Anthropic published a "global workspace" interpretability method for verbalizable internal representations. Nemotron-Labs-Diffusion introduced a tri-mode language model unifying autoregressive, diffusion, and self-speculation decoding, and Tencent released Hy3, a 295B parameter open-source MoE model.

More from AGI Musings