White House blocks GPT-5.6 release

TechCrunch

The White House is requiring OpenAI to conduct a limited, government-approved release of GPT 5.6 before any public rollout, marking a significant shift toward federal oversight of powerful AI models amid cybersecurity concerns. This approach mirrors Anthropic's restricted release strategy for its Claude Mythos model.

Every

OpenAI's GPT-5.5, featuring a new pre-training foundation, delivers a faster, more reliable workhorse model that excels at professional coding and knowledge work, directly competing with Anthropic's Claude in enterprise applications. The model significantly outperforms on engineering benchmarks while offering improved speed and consistency for everyday tasks, though it trails Opus 4.7 in design work and visual composition.

Liquid AI

Liquid AI launched LFM2.5-230M, a 230-million parameter model optimized for ultra-low latency edge applications like robotics and e-commerce, demonstrating agentic capabilities on resource-constrained devices including the Unitree G1 robot.

Hugging Face

Hugging Face launched a single-command solution to deploy vLLM servers on their Jobs infrastructure with per-second billing, OpenAI-compatible APIs, and support for models from 4B to 122B+ parameters. This provides developers flexible infrastructure for testing and evaluation, complementing their production-focused Inference Endpoints service.

Liquid AI

Liquid AI released LFM2.5-230M, a compact 230M-parameter open-weight model that outperforms larger models on tool use and data extraction tasks while achieving exceptionally fast inference on edge devices from Raspberry Pi to smartphones. The model has been demonstrated running on-device on humanoid robots for natural language control, showcasing practical deployment in resource-constrained environments.