May 2025 Summaries

OpenRouter has introduced several new features and enhancements, including streaming reasoning summaries for OpenAI’s o3 and o4-mini models via the OpenRouter Chat Completions API and the ability to submit optional end-user IDs with requests for improved moderation and abuse prevention. They have also added the functionality to generate crypto invoices directly from the credits page for easy payment, and a setting to restrict API usage to a specific third-party key. Additional updates include support for two-factor authentication and Passkeys for enhanced security, as well as AI SDK updates that now include usage accounting and PDF parsing support. These updates come as OpenRouter is working on expanding the capacity of Claude 4 and Gemini 2.5.

OpenRouter has introduced several new features to enhance user experience and security, including the ability to generate crypto invoices, submit end-user IDs, and require third-party API keys. The platform now supports streaming reasoning summaries via the OpenAI o3 and o4-mini models using the OpenRouter Chat Completions API, allowing for more efficient data processing. Users can now include optional user fields with each request to improve moderation, prevent abuse, and enable per-user usage tracking. Additionally, OpenRouter has implemented two-factor authentication and Passkeys for enhanced security and updated its AI SDK to include usage accounting and PDF parsing support. These updates are part of an ongoing effort to expand the capacity of Claude 4 and Gemini 2.5 models.

OpenRouter has introduced several updates, including the launch of passkeys as a passwordless login option to enhance account security. The platform has also improved the developer experience by adding slugs for providers and quantizations, simplifying API targeting, and introducing a "Copy" button for easy access to API calls with slug information. Additionally, a new dropdown feature on documentation pages facilitates quicker navigation and reuse. A new guide on writing TypeScript agents by Mastra is now available in the documentation, offering valuable insights for developers working with TypeScript.

OpenRouter has introduced several updates to enhance user experience and security, including the implementation of passkeys, a passwordless login method recommended for safer account access. The platform has improved developer experience by assigning slugs to providers and quantizations for easier API targeting, with a convenient copy feature for generating exact API calls. Additionally, a new TypeScript Agent Guide by Mastra has been released, offering detailed instructions for writing agents in TypeScript, accessible through their documentation. These updates aim to streamline navigation, improve security, and provide valuable resources to developers.

Cerebras has introduced a new, high-speed computing provider capable of processing over 3,000 transactions per second (TPS) on the Llama 4 Scout model and more than 1,800 TPS on the Llama 3.3 70B Instruct model, offering instant animation generation at a cost of approximately $0.001. The Cerebras chips, known as the largest ever built, feature up to 4 trillion transistors on a single wafer and 40 GB of on-chip memory per wafer, eliminating bottlenecks associated with external memory.

Cerebras has introduced a new high-speed provider capable of delivering over 3,000 transactions per second (TPS) on Llama 4 Scout and more than 1,800 TPS on Llama 3.3 70B Instruct, offering rapid animation generation at a cost of approximately $0.001. The provider utilizes Cerebras chips, which are notable for being the largest ever built, featuring up to 4 trillion transistors and 40 GB of on-chip memory per wafer, thereby eliminating bottlenecks associated with external memory.

OpenRouter has introduced several new features aimed at enhancing user experience and developer capabilities. The enhanced activity page now includes multiple new charts for deeper insights into model usage, allowing users to view personalized model rankings. Improvements in reasoning model performance metrics have been made, with latency now measuring the time until the first reasoning token, and throughput accounting for both reasoning and output tokens. Developers can utilize the new Request Builder to easily generate request body JSONs, and the Prompt Category API allows users to request models optimized for specific prompt categories, such as programming. These updates aim to provide better insights, faster metrics, and more powerful developer tools.

OpenRouter's latest update introduces several enhancements, including an improved activity page with new charts for deeper insights into model usage and personalized rankings. The performance metrics for reasoning models have been refined to measure latency as the time until the first reasoning token and throughput as including both reasoning and output tokens. Developers can now utilize a new Request Builder to generate JSON request bodies with ease, and a Prompt Category API allows for the direct request of models optimized for specific prompt categories, such as programming. These updates aim to provide better insights, faster metrics, and new tools for developers to enhance their experience and efficiency with OpenRouter's offerings.