GPT-5.6 Sol Ultra is set to be included in Codex.
This integration signifies a notable update, as Sol Ultra becomes part of the Codex framework. The specifics of how it will function within Codex or any changes to its capabilities have not been detailed yet. The announcement has garnered significant attention, with the related discussion accumulating hundreds of comments and points online.
Wafer.ai demonstrated GLM5.2 running on AMD MI355X, achieving 2626 tokens per second per node and 213 tokens per second in a single stream at more than twice the cost efficiency compared to Blackwell.
The AMD MI355X GPUs are approximately 2. 75 times cheaper on average than NVIDIA's B300, offering similar hardware specifications.
The Safari MCP server was introduced in Safari Technology Preview 247 as a Model Context Protocol server for web developers.
This server allows any MCP-compatible client to connect and enables agents to understand how code renders in a Safari browser window. It provides access to the DOM, network requests, screenshots, and console output, streamlining the debugging process. Developers can debug directly from the terminal without switching between windows, enhancing efficiency by reducing repetitive actions in the debugging workflow.
GitHub utilized secret scanning to manage over 20,000 alerts across 15,000 repositories and achieve inbox zero.
The process involved distinguishing significant alerts from false positives and developing remediation workflows. This effort was led by Staff Security Engineer Michael Recachinas, who focuses on large-scale security programs and secure development lifecycle tooling. The initiative successfully reached inbox zero within nine months.
Japan's top court has ruled that artificial intelligence cannot be listed as an inventor on patent applications.
This decision aligns with similar rulings in other jurisdictions, reinforcing the notion that only humans can be recognized as inventors. The ruling comes amid ongoing global discussions about the role of AI in the creative process and its implications for intellectual property law. This case highlights the evolving legal landscape as technology continues to advance.
Kimi K2.7 Code, an open-weight model, is now generally available in GitHub Copilot as a selectable option.
This is the first open-weight model offered in the Copilot model picker, providing more choice and a lower-cost option for coding workflows. Kimi K2.
Claude Fable 5 has been re-deployed globally, featuring enhanced cybersecurity safeguards and a new AI jailbreak severity framework.
The cybersecurity safeguards include safety classifiers designed to detect and block harmful cybersecurity uses. The framework for AI jailbreak severity, developed with Glasswing partners, categorizes jailbreaks based on the extent of behaviors they unblock, from minor to highly dangerous.
This model can autonomously make plans, utilize tools like browsers and terminals, and operate at a level that previously required larger models. Claude Sonnet 5 nearly matches the performance of Opus 4.
Google released Copybara, a tool for transforming and moving code between repositories.
Copybara is used internally at Google to synchronize code between confidential and public repositories by designating one as the authoritative source. It supports contributions from any repository, and changes from non-authoritative repositories are transformed and moved to the authoritative one.
GitHub has outlined six security settings that maintainers should enable to strengthen their projects against attacks.
These settings are designed to close easy vulnerabilities, making projects more challenging to compromise. While not guaranteeing complete security, they represent a significant improvement over previous configurations. The guidance comes from Joseph, a prominent cybersecurity expert with extensive experience in educating developers on secure coding practices. His open-source initiatives and educational content have reached thousands globally, emphasizing the importance of proactive security measures.