AI News Brief

1. GLM-5.2 Sparks New "DeepSeek Moment" - Open-Weight Model Beats GPT-5.5 at a Fraction of the Cost

Zhipu AI (Z.ai) released GLM-5.2, a 753B-parameter open-weight model with an MIT license and 1M-token context window. It scores 54.7 on Humanity's Last Exam (ahead of GPT-5.5's 52.2), reached #2 on Code Arena's frontend leaderboard outperforming Claude Opus 4.7, and costs $1.40/M input tokens and $4.40/M output tokens - roughly a sixth of what Claude Opus 4.8 charges. Snowflake's CEO confirmed it is competitive with Opus 4.7 at a fraction of the cost.

2. White House Asks OpenAI to Stagger GPT-5.6 Release Over National Security Concerns

The Trump administration formally asked OpenAI to release GPT-5.6 only to a small group of government-approved partners before any broader public launch, citing national security concerns. This marks the first time the U.S. government has preemptively asked an AI company to restrict a model release. An internal memo indicates the model will undergo a government-controlled preview before wider availability.

3. Anthropic Accuses Alibaba of Illicitly Extracting Claude Capabilities via Distillation

Anthropic alleges that Alibaba used fraudulent accounts to access data from its Claude AI model, sparking a debate over AI distillation - the technique of training a new model using outputs from a more powerful one. GLM-5.2 from Zhipu AI was released a day after the alleged extraction, raising questions about the provenance of training data for Chinese open-weight models.

4. OpenAI and Broadcom Debut "Jalapeno" - First Custom AI Inference Chip

OpenAI unveiled its first custom AI chip, "Jalapeno," built with Broadcom and Celestica specifically for large-scale inference workloads. The chip measures roughly 840mm2 in die size (near the EUV reticle limit) and was developed in a nine-month cycle. OpenAI plans to use enough homegrown chips to consume ten gigawatts of electricity, a direct strategic move to reduce dependence on Nvidia.

5. Qualcomm Unveils AI Data Center Platform, Lands Meta as Customer to Challenge Nvidia

Qualcomm announced new AI data center processors and secured Meta Platforms as a customer for its data center CPUs, marking a significant push into the AI infrastructure market dominated by Nvidia. The expansion includes a full AI infrastructure strategy spanning inference accelerators and server processors.

6. Italy's Domyn Announces Fully Open-Source Frontier AI Model Within a Year

Italian AI company Domyn announced plans to release a fully open-source frontier AI model within a year, aiming to build one of the most advanced open-source systems available. CEO Uljan Sharka made the announcement on Thursday, adding another European entrant to the open-source frontier model race alongside Mistral and others.

7. Japan Anchors Growth Strategy Around JPY101.6 Trillion AI Chip Investment

Japan's government is putting a JPY101.6 trillion ($700B+) AI and semiconductor investment push at the heart of a JPY370 trillion national growth roadmap. Prime Minister Sanae Takaichi is using state-backed investment as a strategic policy lever, targeting domestic chip development and AI infrastructure to reduce reliance on foreign technology.