April 2025 Summaries

OpenRouter has introduced Universal PDF Support, enabling compatibility with any Large Language Model (LLM) available on its platform and handling all types of PDFs, including image-only scans. This feature automatically utilizes native PDF support for providers like Gemini, Anthropic, and OpenAI and transmits embedded images from PDFs directly to multimodal models. Users can enhance functionality by combining it with web search or Auto Router, and it is accessible via API or directly from the OpenRouter Chatroom, supporting multiple PDFs and models in a single request. The service offers two PDF processing engines: mistral-ocr, which provides OCR support with text and embedded images extraction for $2 per 1000 pages, and pdf-text, which offers free text-only extraction without OCR or image support. Comprehensive documentation and implementation guidance are available online, alongside a video demo for more detailed insight.

Universal PDF Support on OpenRouter, as of April 2025, offers a versatile solution for handling various types of PDFs, including image-only scans, and is compatible with any large language model (LLM) available on the platform. It automatically employs native PDF support for providers like Gemini, Anthropic, and OpenAI and can send embedded images directly to multimodal models. The service can be enhanced with web search or Auto Router features and is accessible via API or the OpenRouter Chatroom, supporting multiple PDFs and models in a single request. Users can choose between two PDF processing engines: the default mistral-ocr, which provides OCR support with text and embedded images extraction for $2 per 1000 pages, and the free pdf-text option for text-only extraction. Comprehensive documentation and a video demo are available online for further guidance on implementation and features.

OpenRouter has introduced three quality-of-life updates to enhance user experience and transparency, featuring Activity Chart Filtering, SVG Previews, and In-Stream Usage Accounting. Users can now filter activity charts by model, API key, or provider for more precise trend analysis, while SVG files display inline in chatrooms, allowing quick reviews without downloads. Additionally, developers gain real-time insights into token usage and costs within streamed responses by enabling a specific request setting, thereby eliminating the need for extra API calls.

OpenRouter has introduced three updates to enhance user experience and transparency, focusing on activity chart filtering, inline SVG previews, and in-stream usage accounting. The activity chart now allows users to filter data by model, API key, or provider directly on the Activity page, offering an intuitive way to analyze trends and patterns. SVG files can be previewed inline within chatrooms, enabling users to view diagrams and vector designs without needing to download them. Additionally, developers can receive real-time data on token usage and costs directly in streamed responses by setting a specific parameter in their requests, eliminating the need for extra API calls. These updates aim to improve the overall usability and efficiency for users interacting with OpenRouter's platform.

OpenAI has announced the general availability of GPT-4.1, an advanced language model offering a 1 million token context, alongside its Mini and Nano versions. Pricing for GPT-4.1 is set at $2.00 per input, $0.50 for cached input, and $8.00 for output per 1 million tokens, while the Mini and Nano versions are priced more affordably. The company has also retired the free alpha endpoints for Quasar and Optimus, with API requests now returning a 404 error to prevent unexpected costs. Further information and announcements are accessible through links provided in the announcement.

Quasar Alpha and Optimus Alpha have introduced OpenRouter, featuring the general availability of OpenAI's GPT-4.1, which boasts a 1 million token context. The release includes different versions such as the full-scale GPT-4.1, along with the Mini and Nano versions, each with varied pricing structures based on input and output tokens. The cost for GPT-4.1 is set at $2.00 per input, with reduced rates for cached inputs, and $8.00 for output per million tokens, while the Mini and Nano versions offer more economical options. An announcement was made on the platform X, stating that the free alpha endpoints for Optimus and Quasar have been retired, and API requests to these endpoints will now return a 404 error to avoid any unexpected charges.

Optimus Alpha is a newly introduced general-purpose foundation model currently in its preview phase, optimized for a wide range of tasks with a particular aptitude for coding, and featuring a context length of 1 million tokens. It is available for free during its stealth period, encouraging users to join discussions on Discord to help refine its performance and contribute to future developments. While all prompts and completions are logged by the model lab for enhancement purposes, they are not logged by OpenRouter. Users are invited to try the model through a provided link.

Optimus Alpha is a general-purpose foundation model designed for various tasks with a particular strength in coding, featuring a 1 million token context length. Currently available for free during its stealth period, the model invites users to join discussions on Discord to contribute to its development and performance refinement. It is important to note that while the model lab logs all prompts and completions for improvement purposes, OpenRouter does not. Users are encouraged to try out the model via a provided link.

Quasar Alpha OpenRouter is a prerelease version of an upcoming long-context foundation model, optimized specifically for coding, and is available for free to all OpenRouter users. This initiative marks the first time users are invited to test a foundation model before its public release, with the aim of gathering feedback to enhance the model's performance and guide future developments. Users are encouraged to provide their feedback through the Quasar Alpha thread on Discord, while being informed that all prompts and completions will be logged to help the model lab understand its usage and areas for improvement.

Quasar Alpha OpenRouter is a prerelease of a forthcoming long-context foundation model optimized for coding, featuring a capacity of 1 million tokens, available for free to OpenRouter users. This trial access provides an opportunity to test the model before its official public release and encourages users to give feedback in the Quasar Alpha Discord thread, which will aid in refining the model and shaping future developments. The platform notes that all prompts and completions will be logged to help both the creators and the lab understand user interactions and identify areas for improvement, inviting users to explore its capabilities and share their insights.