China's Moonshot AI released Kimi K2.5, an open source multimodal model with agent swarm technology that enables up to 100 sub-agents to work in parallel, alongside Kimi Code, a coding tool that rivals Anthropic's Claude Code.
Arcee AI, a 30-person US startup, released Trinity Large, a 400B-parameter open source model that rivals Meta's Llama 4 Maverick, addressing concerns about China's dominance in open-weight models and uncertainty around US companies' commitment to open source AI.
A new benchmark called APEX-Agents reveals that leading AI models still cannot reliably perform complex tasks requiring context-switching and multi-domain reasoning. APEX-Agents tests performance on tasks simulating real-world scenarios, with no tested model surpassing 24% accuracy in the benchmark.
Meta is pausing teen access to AI characters globally across its apps while developing a safer version with parental controls, amid mounting legal pressure over child safety concerns.
Symbolic.ai has partnered with News Corp to deploy its AI publishing platform in newsrooms, starting with Dow Jones Newswires, where early testing showed up to 90% productivity gains for complex research tasks.
Comments