GLM 5.2

Security researchers split over how urgently to confront Anthropic’s AI-worm threat

New expert reactions highlight a practical dispute: prepare now for adaptable malware, or prioritize threats beyond a slow, simplified demonstration.

Reflection releases Beam’s open weights, betting on fewer tokens per task

The coding-focused model carries an Apache 2.0 license. Its efficiency pitch rests on company estimates that exclude prompt-reading work—not measured deployment savings.

Reflection Introduces Beam With Early Access Ahead of October Weight Release

The 501-billion-parameter model targets coding and tool-using work. Its efficiency claim is an estimate, not a measured deployment-cost saving.

Perplexity Finds 21% Fewer Tool Failures Between Trained Agent Versions

The live test measured a significant drop in failed tool calls. It found no significant change in strong user dissatisfaction.

Audit Finds No SWE-bench Pro Leaderboard Score Has a Complete Public Receipt

Mozilla Finds Chinese Open Models Are Within 4.4 Months of U.S. Frontier AI

Mozilla’s latest analysis argues that the expensive frontier-model premium is increasingly limited to difficult, long-running work—while the market for open models raises a new concentration problem.

Broadcom Validates Five AI Model Families for VMware’s On-Premises Model Service

The update gives enterprises a path to run models from NVIDIA, Google, NEC, Alibaba Cloud and Z.ai inside their own infrastructure, with new sharing, routing and code-isolation services around them.

Nvidia Optimizes DeepSeek and Qwen as It Warns U.S. Limits Could Hit Business

The company is making its stack easier to use with prominent Chinese open models, but the potential limits it flagged have not been defined as a final rule.