wednesday, june 24, 2026
top 4 trending●previous 24h●generated 42d ago
GLM-5.2, an open-source model from a Chinese AI lab, launched in mid-June 2026 and rapidly gained attention as a frontier-class open model, with users reporting it outperforms or matches proprietary models like Claude Opus and GPT-5.5 on benchmarks including web design, coding, and agentic tasks. The release marks a significant shift in open-model capabilities, particularly for agent-based applications, and has spawned multiple implementations including high-throughput APIs and local deployment guides.
- GLM-5.2 achieved 280 tokens-per-second throughput in third-party API implementations
- Model runs on 64GB Mac hardware and Mac Studio, enabling local deployment
- Outperformed GPT-5.5 on hallucination metrics and Fable 5 on website design tasks
- Positioned as enabling agentic capabilities at frontier level in open-source, compared to 'DeepSeek moment' for agents
- TMax RL training recipe paper released alongside model, establishing open terminal-agent training methodology
Anthropic has accused Alibaba of illicitly accessing and extracting capabilities from Claude AI models without authorization. The accusation centers on unauthorized extraction of Claude's model capabilities.
- Anthropic filed formal accusation against Alibaba for illicit Claude model access
- Allegation involves unauthorized extraction of Claude AI model capabilities
- Incident reported publicly on June 24-25, 2026
Google DeepMind and A24, the independent film and entertainment company, announced a research partnership to explore AI applications in creative and media production. The collaboration marks a rare alignment between a major AI lab and a prominent entertainment studio.
- Google DeepMind and A24 launched their first formal research partnership
- Focus areas include AI applications in film, television, and creative production
- A24 is known for independent films and has faced public criticism over the collaboration
- Partnership announced June 25, 2026
OpenAI and Broadcom announced a custom chip designed for large-language-model inference at scale. OpenAI plans to deploy enough of these chips to consume 10 gigawatts of electricity, equivalent to powering millions of households.
- Chip optimized specifically for LLM inference workloads at production scale
- OpenAI deployment targets 10 gigawatts of total electricity consumption
- Broadcom partnering with OpenAI on silicon design and manufacturing
- Announcement made June 24, 2026
- Move reflects intensifying competition in AI chip development amid capacity constraints