March 2025 Summaries

OpenRouter has expanded its Zero Completion Token Insurance to cover every model available on its platform, ensuring users are not charged for responses with no output tokens due to an empty or error finish reason, regardless of any charges incurred by OpenRouter from model providers. This initiative is expected to save users over $18,000 per week, emphasizing OpenRouter's commitment to fairness and reliability. The insurance is automatically enabled for all users, requiring no additional action, and protected requests are easily identifiable with zero cost on the Activity page. As a leading LLM router, OpenRouter is committed to improving quality standards across the industry by collaborating with model providers and continually enhancing the user experience, encouraging feedback and engagement through their Discord community.

OpenRouter has expanded its Zero Completion Token Insurance to cover all models, ensuring users are not charged for responses with no output tokens or error finish reasons, even if OpenRouter incurs charges from model providers. This initiative, expected to save users over $18,000 weekly, automatically applies to all users without requiring any additional setup. Protected requests are easily identifiable on the Activity page, showing zero cost for zero output tokens. As the leading LLM router, OpenRouter is committed to enhancing quality standards and collaborating with providers to build a user-friendly ecosystem. Users are encouraged to provide feedback and participate in discussions on their Discord community.

OpenRouter has unveiled a series of new and advanced models aimed at enhancing research, storytelling, and efficiency in various tasks. The Sonar models, including Sonar Deep Research and Sonar Reasoning Pro, are designed for comprehensive research and sophisticated analysis, with capabilities to handle complex queries and detailed sourcing. Pricing for these models is influenced by the internal search and reasoning steps used. Additionally, new models like Skyfall 36B v2 and Wayfarer Large 70B focus on enhancing creativity and interactive storytelling, while advanced models like Reka Flash 3 and Cohere Command A offer efficient general chat, coding, and reasoning capabilities. The Jamba series introduces multilingual proficiency and structured outputs, while the Gemma 3 models provide fast, efficient, and compact options for multimodal tasks, with the largest being Gemma 3 27B. These models are accessible for use in both chatroom and API settings.

OpenRouter has introduced a range of new models designed for diverse applications, including comprehensive research, advanced reasoning, storytelling, and efficient general tasks. The Sonar series, featuring models like Sonar Deep Research and Sonar Reasoning Pro, is tailored for in-depth searches and complex queries, while the Sonar Pro model offers capabilities for handling multi-step queries. In the realm of creative storytelling, models such as Skyfall 36B v2 and Wayfarer Large 70B are optimized for immersive narratives and interactive fiction. Additionally, advanced models like Reka Flash 3 and Cohere Command A are available for efficient chat, coding, and multilingual tasks. The Jamba 1.6 series provides multilingual proficiency and structured outputs, and the Gemma 3 family includes various models for fast, multimodal applications. Pricing for some models involves additional charges based on internal search and reasoning steps.