1. OpenAI and Microsoft end exclusive cloud partnership
OpenAI and Microsoft have amended their partnership agreement to end Microsoft's exclusive cloud hosting rights. OpenAI can now serve its products to customers across any cloud provider, including AWS and Google Cloud. Microsoft retains a non-exclusive license to OpenAI's intellectual property and models through 2032 and remains the primary cloud partner. The renegotiation also removes the artificial general intelligence clause and caps Microsoft's revenue-share payments, which are now guaranteed only through 2030.
2. Anthropic releases Memory for Claude Managed Agents
Anthropic has launched a persistent memory feature for Claude Managed Agents in public beta. The feature allows AI agents to retain information across sessions and accumulate knowledge without requiring developers to manually update prompts. Memory is implemented as a filesystem-based layer where data is stored as files that can be exported, managed via API, or edited in the Claude Console. The system supports concurrent access by multiple agents and includes detailed audit logs for tracing, rolling back, or redacting information.
3. Google releases Chrome Prompt API for local Gemini Nano
Google has released the Prompt API, allowing developers to send natural language requests directly to the Gemini Nano model running locally in the Chrome browser. The API supports tasks like content summarization, classification, and extraction without requiring server-side inference. It currently operates on desktop versions of Windows, macOS, Linux, and ChromeOS for devices meeting specific RAM and GPU requirements. Developers can manage inference sessions, clone existing sessions for efficiency, and terminate them to free up local system resources.
4. OpenMOSS releases MOSS-Audio foundation model
The OpenMOSS team has released MOSS-Audio, an open-source foundation model that unifies speech, environmental sound, music, and temporal reasoning. The architecture supports audio captioning, time-aware question answering, and complex reasoning over real-world audio inputs. The release includes four model variants, featuring Instruct and Thinking versions optimized for different reasoning tasks.
5. EvanFlow releases TDD plugin for Claude Code
EvanFlow has released a test-driven development (TDD) plugin for the Claude Code agent. The tool enforces an iterative feedback loop that guides the agent through brainstorming, planning, and execution phases with mandatory user approval checkpoints. During the execution phase, it applies a vertical-slice TDD approach, requiring the agent to write a failing test, implement the minimal code, and refactor before proceeding. The plugin is available directly through the Claude Code plugin marketplace.
6. Anthropic tests Bugcrawl feature for Claude Code
Anthropic is testing an experimental Bugcrawl feature within its Claude Code tool. The feature allows developers to scan entire code repositories for bugs and receive automated fix suggestions. It is currently accessible via a dedicated entry in the side navigation and focuses on general code correctness and quality. Anthropic warns that the tool consumes a high number of tokens, making it best suited for smaller repositories during this testing phase.
7. Utilyze releases open-source GPU monitoring tool
Utilyze has released an open-source GPU monitoring tool designed to provide more accurate utilization metrics than standard tools like nvtop or nvidia-smi. Instead of reporting the fraction of time a kernel is running, Utilyze samples hardware performance counters to measure actual compute and memory throughput relative to theoretical hardware limits. The tool also estimates an attainable utilization ceiling for specific workloads.