Anthropic files its S-1.

OpenAI has officially released GPT-6.1 Terra, explicitly positioning the new release as a fast computer-use and vision model. Rather than marketing this as another incremental update in reasoning or text generation, the company is directly targeting multimodal, agentic workflows.

This release marks a significant shift in the race for market dominance. By optimizing for computer-use and vision, OpenAI is moving beyond the chat interface and attempting to own the execution layer on the user's device. Terra is built to see the screen and interact with it, which is exactly what enterprise developers have been trying to build using complex middleware and older models.

OpenAI wins by natively integrating capabilities that previously required third-party workarounds. The immediate losers are the startups that built their entire business models around giving text-based LLMs visual and computer-control capabilities. If GPT-6.1 Terra handles screen interaction natively and quickly, the market for bolt-on vision agents vanishes overnight.

Artificial intelligence has successfully identified vulnerabilities in PHP, the foundational scripting language that currently runs approximately 70% of the web. This discovery highlights the rapidly accelerating capability of AI systems in complex security research and code analysis.

Finding zero-day holes in infrastructure this ubiquitous shows that automated security prowess is a severe double-edged sword for global web infrastructure. While these specific vulnerabilities were caught and can be patched, the fact that an AI mapped out flaws in a decades-old, heavily scrutinized codebase changes the threat model for every server on the internet.

Security teams win a powerful new auditing tool, but the broader web infrastructure is now on notice. If AI can consistently find exploitable vulnerabilities in legacy languages like PHP, malicious actors will inevitably use the same techniques to automate exploitation at scale. The defense window is shrinking, and relying on manual code reviews is no longer a viable security posture.

DeepSeek has officially open-sourced the actual inference engine that it runs on internally. This release bypasses the usual trend of releasing model weights while keeping the deployment architecture secret, handing developers the exact production-grade tools used by the lab itself.

This move threatens to fundamentally democratize advanced AI deployment. The proprietary infrastructure moats built by major API providers rely on the assumption that running frontier models at scale is too difficult for independent developers. By open-sourcing its inference engine, DeepSeek provides the blueprint for running highly optimized, enterprise-grade inference on independent hardware.

Open-source builders and enterprise teams looking to internalize their AI workloads are the clear winners here. The closed-API providers lose their exclusive grip on the execution layer, as the community now has access to the same high-performance routing and serving infrastructure that powers a major lab.