OpenAI’s Codex Goes GA as Anthropic’s Sonnet 4.5 Smashes Benchmarks
New Models and Model Updates
OpenAI Launches Codex (Generally Available): OpenAI’s AI-powered coding assistant, Codex, is now generally available, offering advanced code generation and autocompletion for developers. The announcement coincides with key updates from OpenAI’s DevDay, expanding its accessibility and capabilities.
Anthropic Releases Sonnet 4.5 with Benchmark Improvements: Anthropic’s latest model, Sonnet 4.5, delivers significant performance gains, as evidenced by higher benchmark scores in computer use tasks. The update enhances Claude’s capabilities, particularly for Max Subscribers.
Qwen3-Next Model Validated in llama.cpp: The Qwen3-Next model has been successfully tested within the llama.cpp framework, enabling local deployment and inference. This integration expands options for running advanced models on local hardware.
Developer Tools and Frameworks
OpenAI Introduces Apps in ChatGPT and Apps SDK: OpenAI’s new "Apps in ChatGPT" feature allows third-party applications to integrate directly into the ChatGPT interface, powered by the newly released Apps SDK. This enables developers to build custom, interactive workflows within ChatGPT.
OpenAI Launches AgentKit for Agentic Workflows: AgentKit is a new visual tool from OpenAI designed to simplify the creation and management of AI agents. It lowers the barrier for non-coders to build complex, automated workflows.
FastFlowLM Enables AI Models on AMD Ryzen AI NPUs: FastFlowLM (FLM) is a lightweight runtime that optimizes models like GPT-OSS, Gemma3, and Qwen3 to run exclusively on AMD Ryzen AI NPUs, offering high efficiency (up to 256k context) without GPUs.
Software Engineering and Transparency
Hugging Face Explains How Transformers Avoids Being a Black Box: A new article from Hugging Face details how the Transformers library (1M+ lines of code) maintains transparency and hackability through modular design, config-driven performance, and the "One Model, One File" approach.
Product Launches and Integrations
Anthropic Releases Claude for Chrome (Early Access): Claude for Chrome is now available in early access for Max Subscribers, offering direct browser integration for seamless AI assistance. The launch coincides with the Sonnet 4.5 performance upgrades.