Meta Releases Llama 4: Open Weights, 400B Parameters, and a Free Commercial License

Zhipu's GLM-5.3-Flash model launched on OpenRouter at 7.5 to 25 cents per million tokens (promotional pricing), delivered entirely on Chinese infrastructure. The model scores 57 on Artificial Analysis' intelligence index at roughly nine cents per task, compared to GPT-5.6 Sol at 59 cents and Grok 4.6 at 94 cents, creating significant cost pressure on enterprise AI budgets already strained by unexpected consumption.

Robot developers are moving beyond GPT-2-era language models to build more capable AI systems for robotic control and reasoning. The article signals a maturation in the field where physical robot platforms are now constrained by the limitations of older, smaller language models rather than hardware. This shift reflects growing demand for more sophisticated AI brains that can handle complex robotic tasks beyond what earlier-generation models can support.

Nvidia researchers have developed a technique that uses linear math to transfer key-value caches between different AI models without recomputing conversation history. The method enables enterprises to switch between small and large models mid-session while reducing compute costs and latency by 2.7 to 25 times compared to traditional recomputation, retaining up to 98% accuracy on compatible model pairs.

Ramp, a financial operations platform, has launched Router, an AI model routing service that allows users and companies to access and switch between multiple large language models through a single API. The service abstracts away the complexity of managing different LLM providers, enabling organizations to route requests dynamically across various models. This move positions Ramp to compete in the growing infrastructure layer for AI applications.