GLM-5.2: A New Model Specialized for Long-Horizon Tasks
Zhipu AI has announced GLM-5.2, a new model optimized for long, complex tasks (long-horizon tasks). It has become a hot topic in the LocalLLaMA community, where it is expected to see adoption in agentic workflows.
Some in the community note that "a year ago local models were toy-level, but Gemma, Qwen, GLM, Kimi, and others have now reached practical quality" — a sign that open-source model quality is improving at a remarkable pace.
Hyperscalers' AI Investment Nearing Cash Flow Limits
According to an analysis by Epoch AI, the five giants — Microsoft, Amazon, Alphabet, Meta, and Oracle — are expanding AI infrastructure investment at a pace of roughly 70% per year, while operating cash flow is growing only 23% annually. If this trend continues, capital spending could exceed cash flow as early as Q3 2026.
Some companies are already relying on external financing, raising questions about the sustainability of the AI infrastructure race.
UK Government Publishes "AI Scenarios 2030"
The UK government has released "AI Scenarios 2030: Helping policymakers plan for the future of AI," a document laying out scenarios for AI development through 2030 aimed at policymakers. It examines the societal and economic impact of AI advances across multiple scenarios, and the content is likely to serve as a reference for AI policymaking in other countries.
Poland Invests $11 Million in ElevenLabs — A National AI Strategy
Vinci SA, a Polish state-backed investment firm, has invested $11 million in AI voice company ElevenLabs. The investment is part of the government-led initiative "AI Lab Poland," which aims to build a domestic tech ecosystem and position Poland as a European AI hub.
Enterprises Worry About AI Usage Costs — The "tokenmaxxing" Problem
According to a report by CBC in Canada, AI cost optimization has become a challenge for enterprises. An effort known as "tokenmaxxing," focused on cutting wasteful token consumption, is drawing attention, and a growing movement is re-examining costs from adoption through operations. On Hacker News, discussion has also grown active around "the fatigue of waiting for AI prompts to complete," highlighting the real-world pain points of teams putting AI to work.
Oracle Bans Generative AI Contributions in OpenJDK
The OpenJDK project, which Oracle oversees, has announced a policy banning code contributions generated by AI. Meanwhile, the GraalVM project, also under the Oracle umbrella, accepts generative AI contributions under certain conditions — an interesting case of the same company taking different positions across its projects. Rule-making for generative AI use in open-source communities looks set to become a key point of debate going forward.